JP3739040B2 - Information processing apparatus and information processing method - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an information processing apparatus and the like, and for example, relates to an apparatus that extracts and processes impression expressions from sentence data.
[0002]
[Prior art]
When it is desired to know an image behind a certain object, such as an image behind a certain word or an image held by people with respect to a certain product, it has been conventionally performed by using a web search or the like.
For example, when there is an expression such as “sports car suitable for the word of a human horse”, it may not be understood what kind of sensory image the word “human horse” is used in this expression.
In this way, if you want to know the sensory image behind the term “human-horse” in the field of cars, you can use queries such as “human-horse-one in a car” (search conditions) using Web search. I was investigating.
[0003]
Also, if you want to know a general brand image of a wristwatch of a certain brand, such as “priced”, “color is beautiful”, etc. He carefully read the user's evaluation articles on the wristwatch on the product evaluation electronic bulletin board, or conducted a large-scale questionnaire.
[0004]
[Problems to be solved by the invention]
As described above, when an image behind a certain target (a word, a product, or the like) is examined by Web search, there are the following difficulties.
First, if a query such as “Integration of humans and horses in a car” is directly entered on a Web search site, it will probably not hit anything. Even if it hits, the number of cases will be very small.
[0005]
Next, it is assumed that the following items are hit as a result of searching for “car and horse-horse” as a query.
1. “Untitled. The union of the human and horse ...
2. "Weekly news. Realization of the unification of humans and horses!
3. “Net News. A sports car suitable for the keywords of a human and horse”
4). “Design plan. Create the ground of the human and human body that is the life of this car.”
5. “Tomorrow's adventure. The fun of a sports car in the word of a horse and horse ...”
[0006]
It is necessary to read through these searched contents to determine what kind of image the “human horse unity” is written in these contents, and it takes a considerable time.
In addition, in the query “car and horse-horse”, a sentence that does not contain the direct word “car” may not be hit and may be omitted from the search.
On the other hand, if only “Hima-Han” is used as a query, a lot of data, that is, a lot of data related to “Hima-Han” that is not related to cars will also be hit. Check each of these data to see which is the relevant data. I have to check it one by one.
[0007]
Furthermore, it is assumed that only necessary data is picked up with a lot of trouble. In order to investigate what kind of image the “human-horse unit” is expressed, when an impression expression word or an impression expression word and a noun phrase linked to it are mechanically extracted from them, all of them are extracted. It does not necessarily represent the image of “one horse in a car”.
Even here, it is necessary to manually select what represents the image of a “human horse in a car”. Wasting time Occurs.
[0008]
As described above, the method of acquiring an image behind a certain object by web search or the like requires reading all data of the search result, and it takes time and effort to grasp the image. there were.
In addition, in order to automatically acquire information about the image, when an impression expression phrase is extracted from the acquired data, it is necessary to pick up only the appropriate phrase from it, which requires manpower and a great deal of time. There was a problem .
The impression expression phrase is composed of three elements: a degree adverb, an impression expression word representing an impression assumed by the degree, and a noun phrase strongly associated with the impression expression word.
[0009]
Therefore, an object of the present invention is to provide an information processing apparatus capable of automatically acquiring an expression representing an image generally held for a certain object, as well as Information processing The law Is to provide.
[0010]
[Means for Solving the Problems]
In order to achieve the above object, the present invention provides a method according to claim 1, As a keyword Search object acquisition means for acquiring an impression expression word and a field for searching the impression expression word; Impression expression words as keywords acquired by the search target acquisition means Sentence acquisition means for acquiring one or more sentences, and the acquired sentences Multiple included Using an extraction means for searching impression expression words, extracting the searched impression expression words, and noun phrases related to the impression expression words as a set, and a dictionary that associates noun phrases with fields, Among the noun phrases extracted in the set, noun phrase specifying means for specifying those belonging to the acquired field, and in the extracted set, output the impression expression word that is set with the specified noun phrase And an information processing apparatus comprising the means.
The invention according to claim 2 is characterized in that the output means outputs the impression expression word in a sentence form using the impression expression word or a list form in which the impression expression words are enumerated. The information processing apparatus described in 1. is provided.
In the invention according to claim 3, impression expression word aggregation means for aggregating impression expression words that are set with the extracted noun phrases as nodes, and the collected impression expression words in a predetermined order. Reordering information creating means for creating reordering information for reordering, wherein the output means determines an impression expression word to be output using the created reordering information. An information processing apparatus according to claim 1 or claim 2 is provided.
The invention according to claim 4 further comprises a noun phrase aggregating means for aggregating noun phrases included in the set using thesaurus information. When the noun phrases are aggregated by the noun phrase aggregating means, the noun phrases are aggregated. The information processing apparatus according to claim 1, wherein the impression expression word set together with the noun phrase is also collected.
In the invention according to claim 5, the output means does not output the impression expression words included in the set that are not included in the impression expression word dictionary in which the impression expression words in the acquired field are registered. An information processing apparatus according to any one of claims 1 to 4 is provided.
The invention according to claim 6 further includes similarity acquisition means for acquiring a similarity between impression expression words included in the impression expression word dictionary, and uses the similarity to set the same noun phrase and set. The information processing apparatus according to claim 5, wherein the impression expression words are collected.
In the invention according to claim 7, the extraction means further includes a function of extracting an impression expression word that is not accompanied by a related noun phrase, and the extraction means is an impression expression that is not accompanied by a related noun phrase. When a word is extracted, the impression expression word is searched from a combination of a noun phrase and an impression expression word prepared in advance, and the noun phrase combined with the searched impression expression word is accompanied by the noun phrase. The information processing apparatus according to any one of claims 1 to 6, wherein the information processing apparatus is set as a set with no impression expression word.
In the invention according to claim 8, in the computer comprising search object acquisition means, sentence acquisition means, extraction means, noun phrase identification means, and output means, the search object acquisition means, As a keyword A search object acquisition step of acquiring an impression expression word and a field for searching the impression expression word; and the sentence acquisition means, Impression expression words as keywords acquired in the search target acquisition step A sentence acquisition step for acquiring one or a plurality of sentences, and the extraction means Multiple included Searching impression expression words, extracting the retrieved impression expression words and the noun phrases related to the impression expression words as a set, and the noun phrase specifying means, the noun phrases correspond to the fields Using the attached dictionary, among the noun phrases extracted in the set, a noun phrase specifying step for specifying one belonging to the acquired field, and the output means, in the extracted set, the specified noun phrase There is provided an information processing method characterized by comprising an output step for outputting impression expression words in a set.
[0011]
DETAILED DESCRIPTION OF THE INVENTION
(First embodiment)
Hereinafter, a preferred first embodiment of the present invention will be described in detail with reference to FIGS. 1 to 11.
[Outline of First Embodiment]
For example, when the expression “human horse integrated” is used in the field of cars, there are cases where it is desired to know what image or meaning this expression is used.
In such a case, from the input screen of the image information search apparatus according to the present embodiment, the field (here, “car”) and the search target of image information (here, “human horse”) such as “one in the car” If the words are entered in natural sentences, the image information of the search object such as “accurate handling” and “smooth running” will be output as emotional expressions.
As described above, the image information search device specifically identifies an image held by many people with respect to the search target (word, product, etc.) of the image information. so Output as emotional expression.
The image information retrieval apparatus can also be said to be an apparatus that translates an image behind a certain object (that is, an image generally held for the object) into a specific emotional expression.
[0012]
The image information search device searches for image information according to the following procedure.
First, regarding the search target Web search for sentence data (text) Collect it or collect it by conducting a questionnaire survey in advance.
The collected sentence data can be said to be impression data estimation target data for estimating the impression of the target.
Next, an impression expression word and a noun phrase strongly associated with this are extracted from all (or a part of) the collected sentence data as a set. For example, the expression “strong running” is the impression expression word “strong” When , Noun phrase "run" But It is strongly connected, and is extracted as an impression expression word / noun phrase set of “powerful / running”.
The impression expression words extracted here are candidates for the material for generating the emotional expression that represents the image behind the search target. At this stage, noun phrases and impression expressions that are not related to the search target are used. A set of impression expression words and noun phrases including words is also included.
In the following description, a noun phrase includes a simple noun.
[0013]
Next, using the list of noun phrases prepared for each field, exclude the impression expression / noun phrase set that includes noun phrases that are not related to the field set by the user from the set of impression expressions and noun phrases that were previously extracted. To do.
The noun phrase list is prepared for each field such as a car system and an accommodation facility system, and noun phrases that are considered to be used in the field are collected. By using this noun phrase list, it is possible to filter out a set of noun phrases and impression expressions that are less relevant to the field specified by the user from the previously extracted set of impression expressions and noun phrases.
At this stage, the set of impression words and noun phrases that include noun phrases that are not relevant to the field is excluded, but the set of impression words and noun phrases that still contain impression words that are not relevant to the field remains. ing.
[0014]
Next, using the impression expression list prepared for each field, exclude the set containing impression expressions that are not related to the field set by the user from the set of noun phrases and impression expressions that were left after filtering. To do.
The impression expression word list is prepared for each field, and impression expression words that are considered to be used in the field are collected. By using this, impression expressions can be further filtered, and a set of noun phrases and impression expressions related to the field can be extracted.
Then, a sensory expression representing the image to be searched is generated using the impression expression word / noun phrase set extracted after filtering and presented to the user.
[0015]
[Details of First Embodiment]
FIG. 1 is a diagram illustrating an example of a logical configuration of the image information search apparatus 1 according to the present embodiment.
As will be described later, the image information search apparatus 1 can be configured using, for example, a personal computer (hereinafter referred to as a PC). In addition, the image information search device 1 can be configured as a stand-alone device or can be configured as a server device that can be connected to a client terminal device via a network such as the Internet or a LAN (Local Area Network).
[0016]
The image information search device 1 is roughly composed of a data input unit 2, an image information search unit 3, a search engine 8, a thesaurus dictionary unit 15, an impression expression word dictionary unit 16, and the like.
As will be described later, the data input unit 2 and the image information search unit 3 can be realized in software by executing an image information search program on a CPU (Central Processing Unit), for example.
The search engine 8 can be configured by a search site connected via the Internet, for example, or can be configured to be included in the image information search apparatus 1. The thesaurus dictionary unit 15 and the impression expression word dictionary unit 16 are stored in a storage medium such as a hard disk.
In the present embodiment, the impression expression word is defined as an adjective expressing an impression, an adjective verb, a degree adverb, etc., but is not limited to this, and can be defined by the design of the impression expression word dictionary unit 16. .
[0017]
The data input unit 2 uses a query input unit 5 for acquiring a query created by the user, a field to which a search target belongs from a acquired query, a field / keyword extraction unit 6 for extracting the search target as a keyword, and the extracted keyword. The search engine 8 includes a search unit 7 that executes a search.
[0018]
The query input unit 5 acquires a query input by the user.
FIG. 2 is a diagram showing an example of the query input screen 23. The query input screen 23 is displayed on a display device such as a display. The query input column 21 is a column in which a user inputs a query using a keyboard or the like.
Queries are natural in the field (respectively “car” and “beer”) and search targets (respectively “human horse together” and “clean”, respectively), such as “one-and-one in a car” or “criminal about beer”. It is expressed by sentences.
[0019]
The search button 22 is a button for inputting the query displayed in the query input field 21 to the query input unit 5, and for example, a pointer (not shown) is placed on the button by a mouse operation and a predetermined mouse button is clicked. Can be activated.
[0020]
In this way, the user enters from the query input field 21 to the query input unit 5 a search target (in the example of FIG. 2, the expression “human horse integrated”) whose image information is to be searched and a field to which the search target belongs in a natural sentence. Can be entered.
Instead of the query input field 21, a field input field and a search target input field may be provided, and the field and the search target may be separately input to each.
[0021]
The field / keyword extraction unit 6 acquires a query from the query input unit 5, performs syntax analysis using a syntax analysis unit (not shown), and extracts a field and a keyword (search target) from the query. The parsing is performed by a publicly known method. That is, after a morpheme analysis is performed with reference to a morpheme dictionary (not shown) to obtain a morpheme string, this is parsed with reference to a syntax analysis dictionary (not shown).
The field and keywords may be extracted from the query using a method other than the syntax analysis.
[0022]
More specifically, the field / keyword extraction unit 6 analyzes a syntax structure such as “<keyword> in <field>” and “<keyword> related to <field>” from the query, and corresponds to <field>. The field representing the field is extracted from the location, and the word representing the search target is extracted from the location corresponding to <keyword>. The word extracted as <keyword> is used as a keyword when the next search unit 7 performs a search. On the other hand, the extracted field is used when the thesaurus dictionary unit 15 or the impression expression word dictionary unit 16 is used later.
That is, when the query is “integrated human and horse in a car”, the field / keyword extraction unit 6 extracts “car” as a field and “human and horse integrated” as a keyword.
[0023]
The search unit 7 (FIG. 1) performs a full-text search by the search engine 8 using the keywords extracted by the field / keyword extraction unit 6, and acquires sentence data related to the keywords.
The search unit 7 stores the acquired text data in a storage medium such as a RAM (Random Access Memory) or a hard disk.
FIG. 3 is a diagram showing an example of the sentence data 25 searched by the search engine 8 and the initial set list 26 generated from the sentence data 25. The initial set list 26 will be described later.
The search unit 7 has acquired a large number of text data 1 such as text data 1 “prioritize the running of the horse and horse ...”, text data 2 “the ground of the horse and horse” is, and so on. .
When data in which text data and image data are mixed is searched, text data is extracted from the data.
[0024]
As described above, the data input unit 2 constitutes a text acquisition unit that acquires text data used for analysis.
Note that the configuration of the text acquisition unit is not limited to this. For example, a questionnaire may be conducted on the Web, and the data may be acquired via the Internet, or a medium such as paper. It can also be configured so that the text written in is input from the keyboard.
[0025]
The set extraction unit 10 (FIG. 1) parses all sentence data of the retrieved sentence data 25 and extracts a noun phrase in which the impression expression word and the impression expression word are strongly related as a set. Hereinafter, this set is referred to as an impression expression word / noun phrase set.
Then, the set extraction unit 10 generates an initial set list from the extracted impression expression word / noun phrase set.
As described above, the set extraction unit 10 constitutes an extraction unit that extracts impression expression words and noun phrases associated with the impression expression words based on a dependency relationship.
In addition, you may comprise so that it may extract from some (for example, a thing with a high search order etc.) of all the cases, without extracting from all the cases.
The impression expression word / noun phrase set may be extracted using a method other than the syntax analysis.
[0026]
The initial set list 26 in FIG. 3 shows an example of an impression expression word / noun phrase set extracted from the text data 25.
In the initial set list 26, “firm / handling” which is an impression expression / noun phrase set related to the car field, “feeling / horse” which is an impression expression word / noun phrase set related to the horse racing field, etc. A variety of fields are mixed.
[0027]
This is because the text data searched using the keyword does not specify the search field and therefore includes text data related to other fields. For example, when a search is performed using “human horse integrated” as a keyword, the acquired text data may be related to the car field, or may be related to the horse racing field or the horse riding field.
Furthermore, there may be a case where a single sentence data includes a noun phrase / impression word set related to the field and a noun phrase / impression word set related to another field.
[0028]
Then, the set extraction unit 10 extracts the set of impression expression words and noun phrases as those related to “human horse integrated” even if the impression expression words are not directly related to the keyword “human horse integrated”.
Therefore, as shown in the initial set list 26 of FIG. 3, for example, an impression expression word / noun phrase set related to “human horse integration” in the horse racing field or an impression expression word / noun related to “human horse integration” in the car field. Phrase sets are acquired together.
These sets of impression expression words and noun phrases that are not related to the vehicle field are later removed by the noun phrase confirmation unit 11 and the impression expression word confirmation unit 12.
[0029]
By the way, an image generally held for a certain search target is often represented only by an impression expression word, but is often represented by some noun phrase in the impression expression word.
For example, the image for the search target “Iroha Saka” may be expressed as “steep slope” or “curvy road”. In this case, it is expressed only by impression expressions such as “steep” and “curly”. Can not.
Thus, the search target of image information is often expressed with a noun phrase. Therefore, by extracting together the impression expression word and the noun phrase that is strongly associated with it, this noun phrase can be used as a clue when narrowing down the impression expression word representing the image of the search target.
[0030]
Next, after describing the thesaurus dictionary unit 15, the function of the noun phrase confirmation unit 11 will be described.
FIG. 4A is a diagram schematically showing a logical configuration of the thesaurus dictionary unit 15. The thesaurus dictionary unit 15 is composed of a field-specific thesaurus dictionary specialized in each field, such as a vehicle-based thesaurus dictionary 30, a hot spring-based thesaurus dictionary 31, a beer-based thesaurus dictionary 32, and so on.
[0031]
FIG. 4B is a diagram showing an example of a noun phrase list of the vehicle thesaurus dictionary 30 and the hot spring thesaurus dictionary 31.
The field-specific thesaurus is, for example, “body”, “style”, “driving”, “handling”,... Is registered. Therefore, the thesaurus according to each field can be used as a noun phrase list specialized for the field.
[0032]
The field-specific thesaurus stores the thesaurus information between registered nouns in addition to the noun phrase list. Thesaurus information is information that specifies synonym relations between nouns, relations of broader terms, and the like.
For example, in the vehicle field, “style” and “body” are synonymous, and “handling” and “driving” are synonymous. In the hot spring field, “bathroom” and “bath” are synonymous, and “bath” is a broad term for “large public bath” and “sento”.
[0033]
The noun phrase confirmation unit 11 (FIG. 1) performs the following three processes while referring to the field-specific thesaurus dictionary of the field.
(1) Delete impression expression / noun phrase sets including noun phrases that are not registered in the field-specific thesaurus dictionary.
(2) Aggregation of impression expressions using noun phrases as nodes.
(3) Aggregation of noun phrases based on thesaurus information.
[0034]
Hereinafter, these processes will be described.
The process (1) is to delete the impression expression word / noun phrase set that does not include the noun phrase related to the field extracted by the field / keyword extraction unit 6 from the initial set list.
The noun phrase confirmation unit 11 searches the noun phrases listed in the initial set list in the field-specific thesaurus dictionary specialized in the field, and checks the noun phrases not included in the dictionary. Then, an impression expression word / noun phrase set including noun phrases that are not registered in the field-specific thesaurus dictionary is removed from the initial set list.
[0035]
The initial set list 26 extracted by the set extraction unit 10 may include an impression expression word / noun phrase set that the user does not want, that is, the field does not match.
For example, despite the user's desire for an image of “one horse in a car”, “Meguro Parma (Racing Horse) is fast”, “Cute Horse” or “Many People” clearly have different images.
[0036]
Humans can reasonably determine that nouns such as “Meguro Parma”, “Horse”, “Human”, etc. are nouns that have nothing to do with “Human Horses in a car”, but computer systems understand that I can't.
For example, “Meguro Palma” is not an important noun in the car field, and is not registered in the noun phrase list of the vehicle thesaurus dictionary 30.
[0037]
Therefore, when the noun phrase of the impression expression word / noun phrase set is “Meguro Parma”, the noun phrase confirmation unit 11 finds “Meguro Parma” even if the noun phrase confirmation unit 11 searches the car field system thesaurus dictionary 30. Absent. Thereby, the noun phrase confirmation unit 11 estimates that this impression expression word / noun phrase set has little relation to the vehicle field, and deletes it from the initial set list 26.
[0038]
That is, the image information search device 1 is a vehicle thesaurus dictionary. 30 Then, by utilizing the fact that nouns that are not important in the car field such as “Meguro Parma”, “Horse”, “Human”, etc. are not listed, “Meguro Parma”, “Horse”, “Human”,・ Understand that impression expression and noun phrase sets that contain nouns such as ・ ・ are not subject to analysis related to the car field.
As a result, the image information search apparatus 1 can delete a noun phrase outside the field and a set of impression expression words strongly associated with the noun phrase.
As described above, the noun phrase confirmation unit 11 constitutes a specifying unit that specifies a noun phrase included in a noun phrase list (thesaurus dictionary 15) that defines noun phrases corresponding to a predetermined target such as a car. .
[0039]
The process (2) is to collect impression expression words using noun phrases as nodes in the impression expression word / noun phrase set. The noun phrase confirmation unit 11 adds the frequency when the same impression expression word exists at the time of aggregation.
In the impression expression word aggregation list 35 of FIG. 5, after the noun phrase confirmation unit 11 deletes an impression expression word / noun phrase set that seems to be related to a field other than the car field from the initial set list 26, the noun phrase is regarded as a node. It is a collection of expressions.
Noun phrases are registered along with their frequencies when there are identical noun phrases in the initial set list 26. For example, it can be seen that “handling” has appeared nine times.
In addition, it can be seen that “solid” appears 6 times and “accurate” appears 3 times as an impression expression word strongly associated with the noun phrase “handling”.
In this way, the noun phrase confirmation unit 11 has a function as an impression expression word aggregation unit that aggregates impression expression words using noun phrases as nodes.
Furthermore, the noun phrase confirmation unit 11 also has a function as a rearrangement information creating unit that generates rearrangement information used when rearranging the collected impression expression words (sorting) later.
The rearrangement information is information for ordering impression expression words according to some algorithm. For example, the appearance frequency of impression expression words can be used as this algorithm, and various types of information such as position statistical information between sentences and numerical information related to the registration status in the subjective evaluation expression dictionary can be considered.
In this embodiment, as an example, the appearance frequency of the impression expression word obtained above is used as the rearrangement condition.
[0040]
The process of (3) uses thesaurus information to combine noun phrases that are synonymous in one, or convert a noun phrase in a lower term to a noun phrase that is a higher term, and so on. Are to be aggregated. Here, thesaurus information is information that identifies a thesaurus relationship such as a synonym relationship between noun phrases.
The noun phrase aggregation list 36 of FIG. 5 is an aggregation of the noun phrases of the impression expression word aggregation list 35 using thesaurus information by the noun phrase confirmation unit 11 (FIG. 1).
[0041]
For example, in the vehicle thesaurus dictionary 30, “handling” and “driving” are synonymous. Accordingly, the noun phrase confirmation unit 11 collects the impression expression / noun phrase set including “handling” or “driving” as the noun phrase as “handling + driving”. In this case, the frequencies of the impression expression words appearing in both of them are added.
Since the impression expression word “exact” appears three times in the noun phrase “handling” and four times in the noun phrase “driving”, the number of occurrences of “accurate” after aggregation is seven.
In addition to this, for example, items in the vehicle thesaurus dictionary 30 such as “style” and “body” that have a synonym relation are collected into one.
[0042]
The noun phrase confirmation unit 11 described above aggregates synonyms, but can also be configured to aggregate the lower terms into higher terms or the synonyms into representative terms using thesaurus information.
For example, the noun phrases “large communal bath” and “sento” are converted into their broader terms, the noun phrase “bath”, and synonyms such as “search”, “search”, and “search” are appropriate representatives. It can be aggregated into the word “search”.
As described above, the noun phrase confirmation unit 11 also has a function as a noun phrase aggregation unit that aggregates noun phrases using the thesaurus relation.
[0043]
The processes (1) and (2) can be executed if there is a noun phrase list specialized in the relevant field, and thesaurus information is not necessary. Therefore, the system can be configured to perform only the processes (1) and (2) using a noun phrase list specialized in the field without using a thesaurus dictionary. In this case, the processing of (3), that is, noun phrase aggregation using thesaurus information is not performed.
The processing order of the above processes (1) to (3) is not limited to the above order, and can be performed in any order.
[0044]
Through the processing of the noun phrase confirmation unit 11 described above, the impression expression word / noun phrase set is deleted from the impression expression word / noun phrase set, and the impression expression words are aggregated using the noun phrases, and further thesaurus information Can be used to aggregate noun phrases.
However, there are still improper expressions of impressions that are inappropriate for the image of “human-horse integrated with cars”, such as “Many handling + driving” and “Handling + driving is difficult”. These inappropriate impression expression words are removed by the impression expression word confirmation unit 12.
In this way, even if the noun phrase is suitable for the relevant field, the impression expression word strongly related thereto may not be suitable for the relevant field. It can be said that, for the first time, an impression expression phrase is effective in the field because both the noun phrase and the impression expression word are suitable for the field.
[0045]
The impression expression word confirmation unit 12 (FIG. 1) uses the impression expression word dictionary unit 16 to determine an impression expression word that seems to have little relation to the field input by the user, and deletes it from the noun phrase aggregation list 36. To do.
For example, if there is an impression expression / noun phrase set of “beautiful / bottle” or “easy to make / bottle” in the beverage field, “room” has no meaning in the beverage field, and the noun “bottle” Although it relates to the beverage field itself, whether or not it is easy to make a bottle does not matter in the beverage field, so such an impression expression word needs to be deleted.
The impression expression word confirmation unit 12 uses a field-specific impression expression word dictionary listing important impression expression words in the field in order to determine whether or not the impression expression word is suitable for the field. .
[0046]
FIG. 6A is a diagram schematically illustrating an example of a logical configuration of the impression expression word dictionary unit 16. As shown in the figure, the impression expression word dictionary unit 16 is composed of a field-specific impression expression word dictionary specialized for each specific field, such as “car”, “hot spring”, “beer”,. ing.
In the field-specific impression expression word dictionary, impression expression words considered to be important in the field are listed and registered.
[0047]
FIG. 6B is a diagram showing an example of the vehicle impression expression word dictionary 41. In the car-related impression expression dictionary 41, for example, “firm”, “individual”, “accurate”, “smooth”, etc., impression expression words that are considered important in the car field are included. There are no registered impression expressions that are registered and are not considered important in the car field, such as “Strong”, “Hard”, and so on. Therefore, the impression expression word confirmation unit 12 can determine whether or not the impression expression word is related to the vehicle field by referring to the vehicle system impression expression word dictionary 41.
[0048]
For example, when the impression expression word confirmation unit 12 checks the vehicle field, the impression expression word in the noun phrase aggregation list 36 (FIG. 5) is searched for and registered in the field-specific impression expression word dictionary of the vehicle field. If not, the impression expression word is deleted from the noun phrase aggregation list 36.
[0049]
FIG. 7 is a diagram for explaining a case where the impression expression word confirmation unit 12 deletes an impression expression word from the noun phrase aggregation list 36.
In the noun phrase aggregation list 36, “strongly competitive running” is not “human horse united” related to cars, but “human horse integrated” related to horse racing.
The impression expression word confirmation unit 12 searches the vehicle impression expression word dictionary 41 for “strong game”, and when the impression expression word is not registered in the vehicle expression expression word dictionary 41, the impression expression word is Recognize that it is not related to the field.
[0050]
Similarly, the impression expression word confirmation unit 12 deletes “many” and “hard”, and generates a result list 37 including only the impression expression words related to the vehicle field.
In addition, impression expression word The confirmation unit 12 stores the generated result list 37 in a storage medium such as a hard disk.
As described above, the impression expression word confirmation unit 12 constitutes an impression expression word specifying unit that specifies items not included in the impression expression word list.
In the above, the impression expression word confirmation unit 12 deleted inappropriate impression expression words from the noun phrase aggregation list 36, but the remaining impression expression words are further aggregated by using a subjective evaluation expression dictionary described later. It can also improve the performance.
[0051]
The output unit 13 (FIG. 1) sorts and outputs the contents of the result list 37 by sorting the noun phrases in descending order of frequency and the impression expression words in descending order of frequency. In general, it is considered that noun phrases and impression expressions with higher appearance frequency are more important. Thus, by sorting noun phrases and impression expressions in descending order of frequency, important noun phrases and impression expressions are ranked higher. Can be ranked.
When there are a plurality of impression expression words and noun phrases, it is possible to determine a threshold value as appropriate and output a word exceeding the threshold value.
[0052]
As a criterion for determining the importance, there is a method using not only the appearance frequency but also, for example, the following position statistical information between sentences and numerical information related to the registration status in the subjective evaluation expression dictionary.
Here, as position statistical information between sentences, specifically, in a sentence searched by a search engine, a noun phrase or impression expression word belongs to a specific sentence (for example, the relevant field included in the noun phrase list). This refers to information about how much text is included, such as a sentence in which nouns appear more frequently, such as a sentence having a high appearance frequency and rarely appearing in other documents.
For such noun phrases and impression expressions, even if the total frequency is low, take measures such as increasing the importance of other expressions and placing them higher in the sorting order. be able to.
[0053]
Specifically, numerical information related to the registration status in the subjective evaluation expression dictionary (described later) represents the element expression (axis) of the subjective evaluation expression dictionary when removing unnecessary impression expression words using the subjective evaluation expression dictionary. Impression expression words or other impression expression words registered in the subjective evaluation expression dictionary with similar coordinates so that the importance is increased even if the frequency is low, and the result of sorting is higher. It is also possible to do this. As described above, various methods of determining the importance can be considered in addition to the appearance frequency.
[0054]
FIG. 8A is a diagram illustrating an example of the output screen 47 when the result output by the output unit 13 is displayed on a display device using a display unit (not shown). output Screen 47 Is Query This screen is displayed when a search is executed on the input screen 23 (FIG. 2). As an example, it is assumed that the output unit 13 outputs the top two noun phrases, and outputs the top one of the impression expression words.
output The screen 47 includes an image information display field 45 and a search field 46. In the image information column 45, image information of the search target in the field (for example, “integrated human and horse” in the car field), that is, a sensory expression representing an image behind the search target is displayed.
[0055]
The image information column 45 is composed of two upper and lower stages. The top row shows the most frequently used noun phrase in the result list 37 (13 times for “handling + driving”) and the corresponding impression expression word that has the highest number of occurrences (7 times for “accurate”). ) Is displayed.
[0056]
The lower row shows the result list 37 that has the second highest number of noun phrases ("Run" 13 times ) And the impression expression word corresponding to the largest number of appearances (“smooth” is 7 times) are displayed.
In this way, sorting the results seems to match the image of the majority of people with more frequent nouns and impression expressions, and the frequent nouns and impression expressions are placed behind the words. This is because it seems appropriate to adopt it as a certain sensory expression.
[0057]
The search column 46 is a column for inputting a noun phrase for searching an impression expression word as a key. For example, image information display In the column 45, only “accurate”, which is the most frequently expressed impression expression word for the noun phrase “handling”, is displayed. Therefore, when “handling” is input in the search column and a predetermined operation such as pressing the return key of the keyboard is performed, for example, as shown in the search result column 48 of FIG. ”And the impression expression words“ accurate ”and“ solid ”are also displayed.
[0058]
As described above, the output unit 13 constitutes an output unit that outputs the impression expression word included in the impression expression word / noun phrase set acquired from the text data input from the data input unit 2 in a predetermined format. Then, when outputting, it is possible to determine an impression expression word to be output using the appearance frequency, such as preferentially outputting an impression expression word having a high appearance frequency. Further, the output unit 13 does not output those that are not registered in the impression expression word list by the impression expression word confirmation unit 12 because they have already been deleted.
Also, as shown in FIG. 8, the output format may be a combination of a noun phrase and an impression expression word such as “Handling”, “Accurate”, etc. The impression expression words may be output alone, or the impression expression words registered in the result list may be output in a list format.
[0059]
FIG. 9 is a diagram illustrating an example of the configuration of the image information search apparatus 1. The image information search device 1 can be configured using, for example, a PC.
The image information search apparatus 1 is configured by connecting an input device 55, an output device 56, a communication control unit 57, a storage device 58, a storage medium driving device 61, an input / output interface 62, and the like to the control unit 51 via a bus line 63. Has been.
[0060]
The control unit 51 operates in accordance with an image information search program or the like, and generates a sensual expression representing an image generally held for a search target, controls the entire image information search apparatus 1, and the like.
The control unit 51 includes a CPU 53, a ROM (Read Only Memory) 52, a RAM 54, and the like.
[0061]
The ROM 52 is a read-only memory that stores various programs, data, parameters, and the like for the CPU 53 to perform various calculations and controls. The CPU 53 can read programs, data, parameters, and the like from the ROM 52, but generally does not rewrite or delete them.
[0062]
The RAM 54 is a random access memory used as a working memory by the CPU 53. The CPU 53 can write and erase programs and data in the RAM 54. In the present embodiment, the RAM 54 acquires a query from the user and extracts a field and a keyword from the query, generates an initial set list, an impression expression aggregation list, a noun phrase aggregation list, a result list, or a result. It is possible to secure an area for performing various processes such as outputting.
[0063]
The input device 55 is composed of input devices such as a keyboard and a mouse.
The keyboard is a device for inputting information such as characters and numbers to the image information search device 1. The keyboard is composed of keys for inputting kana and English characters, numeric keys for inputting numbers, various function keys, cursor keys, and other keys. Information can be entered.
The user can input a query to the query input field 21 (FIG. 2) using the keyboard.
[0064]
A mouse is a pointing device. When operating the image information search apparatus 1 using GUI (Graphical User Interface) or the like, it is displayed on the query input screen 23. Search Clicking button 22 to start a search But it can.
[0065]
The output device 56 includes a display device, a printing device, and the like, for example.
The display device is a device for presenting information formed on, for example, a CRT (Cathode Ray Tube) display, a liquid crystal display, a plasma display, or the like on a screen.
The display device can display the query input screen 23, the output screen 47, and the like.
[0066]
The printing apparatus is an apparatus that prints data output from the output unit 13 on a printing medium such as paper. The printing apparatus includes various printer apparatuses such as an ink jet printer, a laser printer, a thermal transfer printer, and a dot printer.
[0067]
The communication control unit 57 Leased line Is a device for connecting the image information search device 1 to a network such as the Internet or a LAN (Local Area Network), and includes a modem, a terminal adapter, and other connection devices.
The communication control unit 57 is controlled by the CPU 53, and can send and receive data to and from a server device or a client terminal device on the network according to a predetermined protocol such as TCP / IP (Transmission Control Protocol / Internet Protocol).
In the present embodiment, the search unit 7 accesses the search site via the communication control unit 57 when using the search engine 8 installed in the search site on the Web.
[0068]
The storage device 58 includes a readable / writable storage medium and a drive device for reading / writing programs and data from / to the storage medium. A hard disk is mainly used as the storage medium, but it can also be configured by other readable / writable storage media such as a magneto-optical disk, a magnetic disk, and a semiconductor memory.
[0069]
The storage device 58 is provided with a program storage unit 59, a data storage unit 60, a storage unit for other programs and data (not shown), and the like.
Program storage unit 5 9 Stores an image information retrieval program, a syntax analysis program, an OS (Operating System), a communication control program, and other programs.
The image information search program is a program for causing the CPU 53 to perform the image information search function. When the program is executed by the CPU 53, the modules of the respective units shown in FIG.
[0070]
The syntax analysis program is a program for causing the CPU 53 to exhibit the morphological analysis function and the syntax analysis function. By using this morphological analysis function, a field and a keyword are extracted from a query input by the user, or an impression is expressed from sentence data. A word / noun phrase set can be extracted.
The OS is a basic program for operating the image information search apparatus 1.
The communication control program is a program for controlling the communication control unit 57. The data storage unit 60 stores text data searched by the search unit 7, a thesaurus dictionary unit 15, an impression expression word dictionary unit 16, a morpheme analysis dictionary used for morpheme analysis, a syntax analysis dictionary used for syntax analysis, and the like. Yes.
[0071]
The CPU 53 is a device that performs various processes performed by the image information search device 1 in accordance with a program stored in the ROM 52, a program stored in the storage device 58, and the like. The program is read (loaded) from the storage device 58 and executed. Or read / write data in the storage device 58, or control each unit of the image information search device 1.
[0072]
The storage medium drive device 61 is a drive device for driving a removable storage medium so that the CPU 53 reads and writes data. Examples of the removable storage medium include a magneto-optical disk, a magnetic disk, a magnetic tape, a semiconductor memory, a paper tape punched with data, and a CD-ROM. Note that CD-ROMs and paper tapes can only be read.
It is also possible to collect sentence data used for image information search in a storage medium in advance and read it from the storage medium driving device 61.
In addition, data output from the output unit 13 can be written to a storage medium using a storage medium driving device.
[0073]
The input / output interface 62 is configured by, for example, a serial interface or another standard interface. By connecting an external device corresponding to the interface to the input / output interface 62, the function of the image information search apparatus 1 can be expanded. Examples of such external devices include a storage device such as a hard disk, a speaker, and a microphone.
[0074]
In the present embodiment, the image information retrieval apparatus 1 is Stand alone Image information retrieval device 1 And the client terminal device are connected to each other via the Internet or the like, and the image information search device 1 can be used as an image information search server device. In this case, the image information search server device transmits data for displaying the query input screen 23 to the client terminal device, and acquires the query input by the user via the network.
Then, the image information search server device generates an emotional expression using the query acquired from the client terminal device, and transmits this to the client terminal device.
[0075]
FIG. 10 is a flowchart for explaining the operation of the image information search apparatus 1.
First, the CPU 53 displays the query input screen 23 on the display. Then, the CPU 53 acquires a query input from the query input screen 23 by the user using a keyboard or the like (step 10).
Next, the CPU 53 parses the acquired query and extracts a field and a keyword (step 20).
[0076]
Next, the CPU 53 searches the text data with the search engine 8 using the extracted keyword (step 30). This search uses a search engine 8 installed outside the image information search apparatus 1, but the search engine 8 may be installed inside the image information search apparatus 1.
Next, the CPU 53 extracts a noun phrase and an impression expression word strongly associated with the noun phrase from the retrieved text data as a set, and generates an initial set list (step 40).
[0077]
Next, the CPU 53 performs the following three processes while referring to the field-specific thesaurus dictionary.
First, the CPU 53 deletes an impression expression word / noun phrase set including a noun phrase that does not correspond to the field set by the user from the initial set list.
Next, the CPU 53 aggregates the impression expression words using the noun phrases as nodes and generates an impression expression word aggregation list.
Furthermore, the CPU 53 refers to the thesaurus information in the field-specific thesaurus dictionary, and adds noun phrases. check Aggregating and generating a noun phrase aggregation list (step 50).
[0078]
Next, while referring to the field-specific impression word dictionary, the CPU 53 selects an impression word that does not correspond to the field set by the user from the noun phrase aggregation list. check The result list is deleted (step 60).
Next, the CPU 53 sorts the contents of the result list appropriately and outputs them as image information search results (step 70).
[0079]
As described above, the image information search apparatus 1 collects text data related to a certain matter (search target), and acquires an impression expression word / noun phrase set therefrom. Then, by using a thesaurus dictionary and an impression expression word dictionary in the corresponding field, it is possible to automatically acquire image information related to the matter.
[0080]
For example, the basic idea is to obtain a general image of people, that is, an image that most people have, using the appearance frequency information in text data, etc. Furthermore, by adding a search function for impression expression words linked to a noun phrase, the user can derive a result that is more appropriate to his / her sensitivity.
[0081]
Further, by changing the field-specific thesaurus dictionary and the field-specific impression expression word dictionary into different fields (themes), different results can be obtained even if the same data is used.
The image of the “human” in the “car” and the human in the “horse racing” should be different. If “horse racing” is the theme, an image relating to “handling” is not output.
[0082]
(Modification of the first embodiment)
As described above, in the present embodiment, the result expression 37 is generated using the impression expression word dictionary unit 16, but instead of the impression expression word dictionary unit 16, the subjective evaluation expression dictionary described in Japanese Patent Application No. 11-281501 is used. It can also be used. When this dictionary is used, impression expression words registered in the result list can be further aggregated.
The subjective evaluation expression dictionary is a dictionary that stores impression expression words typified by adjectives / adjective verbs representing impressions and information on the similarity between impression expression words.
In general, the subjective evaluation expression dictionary is used in fields such as data retrieval. Therefore, configuring the image information retrieval apparatus 1 by diverting an existing subjective evaluation expression dictionary without creating a new impression expression word dictionary unit 16 is useful for reducing manufacturing costs.
[0083]
FIG. 11 is a diagram illustrating an example of a logical configuration of the subjective evaluation expression dictionary.
In the subjective evaluation expression dictionary, impression expression words used for expressing an image, feeling, or impression of a certain object in a natural language are stored for each category (field).
For example, if you are interested in beer, you can use the impression expressions as “rich”, “crisp”, “savory”, “smooth”, “fresh”, “smooth”, “ “Soft”, “Tasteless”, “Smooth”, etc. are registered.
In this way, the impression expression word important for expressing an image or the like with respect to a certain object is stored for each field in the same way as the impression expression word dictionary unit 16. Therefore, the subjective evaluation expression dictionary can be used in place of the impression expression word dictionary unit 16.
[0084]
Further, the subjective evaluation expression dictionary is associated with each other based on the similarity of images represented by the impression expression words. By using this similarity, similar ones of the impression expressions in the result list 37 can be collected.
In the subjective evaluation expression dictionary, the similarity of each impression expression word is associated as follows.
First, a plurality of impression expressions that are considered to best represent the characteristics of the object are selected and used as element expressions (subjective similarity elements). Other impression expression words are regarded as weighted addition of these element expression words. That is, each impression expression word is represented as a vector having each element expression word as a component.
The subjective evaluation expression dictionary stores vector components of each impression expression word.
[0085]
For example, taking beer as an example, “Body”, “Kire”, “Umami”, “Nodokoshi”, “Exhilaration”,... Are selected as element expression words.
The impression expression word “mellow” is expressed in the vector space spanned by these element expression words as follows. (Body, sharpness, deliciousness, ...) = (4, -2, 2, ...).
Any impression expression word is an element expression word Or The vector component of each impression expression word is artificially determined by the dictionary design. decided Is. Therefore, the degree of similarity of impression expression words can be flexibly set by designing the dictionary.
[0086]
There are various methods for aggregating impression expressions using the subjective evaluation expression dictionary. For example, impression expressions that are not element expressions are the most common. It can be configured to be aggregated into element expression words having large components.
In addition, for example, the impression expression word “classy” has both the vector components of the element expression words “koku” and “umami” equal to 3, but when there are a plurality of element expression words having the same component, for example, Priority should be given to the element expression words to be aggregated, such that “rich” is given priority over “umami” and the impression expression word “classy” is gathered into “rich”.
[0087]
Or, as another method, it is also possible to configure so as to aggregate the impression expression word having the shortest Eugrid distance in a vector space spanned by the element expression word, without necessarily collecting the element expression word. A short Eugrid distance means that these two impression expressions are similar.
For example, although specific vector components are not shown, the Eugrid distance between the tips of impression expression words “rich” and “rich” in the subjective evaluation expression dictionary is set to be short. For this reason, these two impression expression words can be summarized as “abundant + rich”.
In this way, by using the subjective evaluation expression dictionary, the impression expression word confirmation unit 12 has a function as a similarity acquisition unit for acquiring the similarity between impression expression words, and the impression expression word is determined based on the acquired similarity. It becomes possible to aggregate.
Note that the above is an example, and impression expressions using other algorithms may be aggregated.
[0088]
In more detail, impression expression words (including element expression words) are further provided with ranges in the components of the vector space spanned by the element expression words.
In this way, by setting the range, it is possible to designate a certain region in the vector space centering on the point at the tip of the vector of the impression expression word and having the range as the spread related to the component. That is, a certain area in the vector space can be specified by specifying the impression expression word.
This area is mainly used when the subjective evaluation expression dictionary is used in the search system.
[0089]
As described above, in the modification of the present embodiment, when the subjective evaluation expression dictionary is already held, it is used in place of the impression expression word dictionary section 16 to reduce the cost of creating the impression expression word dictionary section 16. can do.
Then, by using the similarity between impression expression words stored in the subjective evaluation expression dictionary, the impression expression word confirmation unit 12 has an impression expression word aggregation function, and the impression expression words listed in the result list 37 are displayed. Furthermore, it becomes possible to aggregate.
[0090]
(Second Embodiment)
In the first embodiment, a case has been described in which a general user translates an image into an emotional expression, but in this embodiment, a user's live voice such as a user questionnaire or an article posted on the Web is used. An embodiment will be described in which a user is assumed to be a service provider, customer support, or marketer, not a general user.
More specifically, for example, a brand on an online shopping site by translating what image the user has for each brand such as wristwatch, handbag, perfume, etc. Perform image completion.
[0091]
The system of the present embodiment can be configured with the main part using the image information search unit 3 (FIG. 1) of the first embodiment. For this reason, the same components as those in the first embodiment will be described using the same reference numerals. The hardware configuration is the same as that of the first embodiment (FIG. 9).
In this embodiment, user opinions regarding each brand are collected in advance as text data by a questionnaire or an external search engine such as the Web and stored in the storage device 58 or the like. The following processing is performed on already collected text data.
[0092]
The set extraction unit 10 extracts an impression expression word / noun phrase set for each brand from the collected sentence data and generates an initial set list.
The noun phrase confirmation unit 11 collects impression expression words using the initial set list as a noun phrase as a node. If the same noun phrase or impression expression word appears multiple times, the frequency is added.
The impression expression word aggregation list 65 shown in FIG. 12A is an aggregation of impression expression word / noun phrase sets extracted for a certain brand using noun phrases as nodes. For this brand, the noun phrase “image” appears 15 times in the collected text data, and “advanced” is 8 times as an impression expression word strongly related to this noun phrase, “adult” is 3 times, It can be seen that “settled” appeared four times. At this stage, noun phrases “shopping diary” and “chance” that are not related to the brand image are also extracted.
[0093]
Further, as shown in FIG. 12B, the noun phrase confirmation unit 11 is not registered in the field-specific thesaurus dictionary while referring to the field-specific thesaurus dictionary of the field (that is, related to the field of the brand). A list of impression expressions combined with noun phrases “shopping diary”, “chance” and impression expressions connected to this. 65 And synonymous noun phrases (“price” + “price” and “coloring” + “color”) are aggregated to generate a noun phrase aggregation list 66.
By using the field-specific thesaurus dictionary, it is possible to automatically delete impression expression phrases acquired from Web advertisements such as “big opportunities” and other asided sentence text data.
[0094]
The impression expression word confirmation unit 12 uses a subjective evaluation expression dictionary of the brand field (in this embodiment, the subjective expression expression dictionary is used instead of the impression expression word dictionary unit 16 to collect impression expression words). Then, it is checked whether or not the impression expression word in FIG. 12B exists in the subjective evaluation expression dictionary related to the field. If it does not exist, delete it and make the frequency of the noun phrase correspond to it.
Furthermore, since the subjective evaluation expression dictionary is used as the impression expression word list, almost synonymous impression expression words having very close meanings such as “rich” and “rich” can be aggregated.
In this way, the impression expression word confirmation unit 12 generates the result list 67 shown in FIG.
[0095]
Moreover, in this Embodiment, since a user's raw voice is used as text data, the expression which is not assumed may appear. Therefore, an impression expression word list that is unnecessary, that is, not recognized as an image is created as an impression expression word list (called a stop word list), and the impression expression words registered in the stop word list are written as text data. If it is extracted from the file, it may be configured to delete it. In this case, for example, if it is determined that the image “light color” is unnecessary as a brand image, “light” may be registered in the stop word list.
[0096]
The output unit 13 outputs the result list 67 to a storage medium or the like for each brand field and each brand type. In this way, for example, an image for each brand such as “unusual design” for the brand A of the wristwatch and “skin is excellent” for the brand B of the handbag is obtained.
Or it can total for every company, such as A company and B company, or it can total for every individual goods even if it is the same brand.
[0097]
As the situation (case) in which the brand image obtained by the above procedure is used, for example, the following three types are conceivable.
(1) Display the brand image directly.
(2) Aggregate images of brands for each user attribute (age, gender, etc.).
(3) Present the time change of the brand image.
Next, these utilization methods will be described.
[0098]
In the case of (1), when the user designates a brand, a sensuous expression representing the brand image is displayed and the brand image is directly used.
FIG. 13 is a diagram showing an example of the brand image screen 71 displayed on the display constituting the output device 56 (FIG. 9). This screen shows a state in which a user narrows down an online shopping site with a wristwatch as a target field.
The brand image screen 71 presents the brand images obtained from the result list 67 so that the user can easily see them.
[0099]
The brand image screen 71 includes a brand selection area 72, an image information display area 73, an image information search field 74, a search button 75, and the like. In the brand selection area 72, brand names belonging to the relevant field are displayed as “Bulgariya”, “Cartel”,. Each brand name is provided with a radio button, and when any one of these radio buttons is clicked by a mouse operation, the button can be selected. At this time, if there is another button in the selected state, the selected state is released. In the figure, “Bulgaria” is selected.
[0100]
In the image information display area 73, for the brand selected in the brand selection area 72, the result list shown in FIG. 67 The brand image obtained from is displayed by sensibility expression.
Image information Display area 73 Then, the image information such as “luxury image”, “price is high”, “color is abundant and outrageous” behind the noun phrase of each brand name is automatically displayed.
That is, the image information display area 73 can automatically extract and display an image held by a larger number of people, as in the first embodiment.
[0101]
The image information search column 74 is a column for inputting an image for searching for a brand. For example, when image information such as “brand with abundant colors” is entered in the image information search field 74 and the search button 75 is clicked, the result list is not shown. 67 A list of brands that have image information such as “abundant colors” is displayed.
[0102]
Furthermore, using the image behind the obtained brand, it is possible to visually display the difference in image for each company as shown in FIG.
The pie chart shown in FIG. 14 is created using the brand image of each company obtained from the result list and the annual sales of each company. From the pie chart, it can be seen that the user mainly has an image that Company A is friendly, Company B is stylish, and Company C is rich in color.
This is an example, and the brand image can be analyzed from various viewpoints and used for market analysis.
[0103]
Next, a case will be described in which images for the brand of (2) are tabulated for each objective attribute (age, sex, occupation, residential area, etc.) of the user.
In this case, first, a user questionnaire regarding the company brand (or another company brand) is conducted. Then, when conducting the questionnaire, the user is asked to fill in the attributes, and when extracting the impression expression word / noun phrase set from the questionnaire results, the attributes of the user who filled in the questionnaire as the extraction source are associated with each other. That is, when a certain impression expression word / noun phrase set is specified, it is possible to specify the attribute of the questionnaire writer who is the extraction source.
[0104]
The result list 67 of FIG. 12C is created using the impression expression word / noun phrase set thus obtained.
After that, the ratio is calculated for each age, gender, and occupation using attributes. At that time, if you collect impression expressions using common nouns such as “atmosphere” in the image behind the brand as a node, you can see what kind of image you have for your brand's atmosphere by age it can
[0105]
In the above example, it was necessary to associate the impression expression word / noun phrase set with the user who filled out the questionnaire. For example, even if there is no association mechanism, processing for each attribute can be performed.
[0106]
Fig. 15 (a) shows the image of the company brand for each user age group. In As a result of analysis, the image with the largest percentage is shown in the table.
Users in their 30s feel that this company has an “friendly atmosphere”, and users in their 20s feel that “there is an inaccessible atmosphere”. Furthermore, it turns out that the user of teenagers has an image of "price is too high".
[0107]
FIG. 15B is a table in which impression expression words are collected using the atmosphere as a node, and the words are tabulated according to the age of the user.
For the noun phrase “atmosphere”, impression expressions such as “friendly”, “inaccessible”, “too flashy”,... Are listed. From the table, it can be seen that the number of users who have the image that “atmosphere” is “friendly” is 0% in their teens, 50% in their 20s, and 80% in 30 units. You can also find out the composition by age group for other impression expressions.
Further, for example, the noun phrase “price” can be analyzed in the same manner. In this case, image information (noun phrases / impression expression words) including the word “price” may be collected.
[0108]
As described above, in this example, the image behind the obtained brand can be analyzed by paying more attention to the general noun.
In addition, it can be applied to various image analysis such as brand images of other companies, brand images of certain products, and images of actors and talents.
[0109]
Next, the case where the time change of the brand image (3) is presented will be described.
FIG. 16 is a diagram showing changes in the company brand over the past four years.
This is the result of extracting the image of the company brand from the user questionnaire collected over the past four years, creating a list similar to the result list, and comparing the three elements “luxury image” and “priced” ”And“ Abundant colors ”are graphed.
It is not possible to know “what” is abundant only by “abundant”, but it is possible to recognize that “coloring is abundant” by acquiring a noun together with image acquisition.
Data defining these figures is output from the output unit 13 (FIG. 1), and can be printed out from a printer connected as the output device 56 (FIG. 9) or displayed on a display.
[0110]
From FIG. 16, it can be seen that the image of the company's brand has changed greatly over the four years from 1998 to 2001. From 1998 to 2001, it can be seen that “colorfulness” is abruptly decreasing and the ratio of “luxury image” is increasing.
When determining management strategies and sales policies, this information on time-varying brand image is important reference data.
In addition, as an example, we described the time change of the image of our own brand, but it is not limited to this. Analyzing the time change of various images such as the brand image of a specific product or the image of a tourist spot. be able to.
[0111]
(Third embodiment)
In the present embodiment, product image supplementation is performed when a user performs Web shopping or selects a product at a store.
As an example, (1) the case of supporting the purchase of a music CD (Compact Disk) and (2) the case of supporting the total coordination of a party will be described.
Hereinafter, portions corresponding to the components of the first embodiment will be described using the same reference numerals.
[0112]
(1) Music CD purchase support
In the present embodiment, for example, CD purchase at a store is supported using a CD purchase support apparatus.
The main part of the CD purchase support apparatus can be configured using the image information search unit 3 (FIG. 1) of the first embodiment. Therefore, the same constituent elements as those in the first embodiment will be described using the same reference numerals. The hardware configuration is the same as that of the first embodiment (FIG. 9).
[0113]
First, search is performed by a search engine using a song name as a key, and sentence data is collected for each song name. The collected text data is stored in the storage device 58 or the like. The subsequent processing is performed on the stored text data.
FIG. 17A is a diagram showing an example of collected sentence data. As shown in the figure, text data A, B,... Are prepared for songs A, B,. In the figure, only songs A and B are shown.
In this embodiment, one sentence data is prepared for each song as an example. However, the present invention is not limited to this, and a plurality of sentence data is collected for each song as in the first embodiment. In addition, the appearance frequency of noun phrases and impression expressions may be added.
[0114]
For the collected sentence data, the set extraction unit 10 extracts an impression expression word / noun phrase set from the sentence data for each song.
FIG. 17B is a diagram showing an example of noun phrases / impression expression words extracted from sentence data for music A and music B. FIG. For example, an impression expression word / noun phrase set such as “uptempo / rhythm” or “cool / singer” is extracted from the sentence data related to the song A.
[0115]
The noun phrase confirming unit 11 deletes nouns that are not registered in the music thesaurus dictionary 85 shown in FIG. 17C or aggregates synonymous noun phrases. Thereby, for example, the singer is concentrated on the singer.
The impression expression word confirming unit 12 uses the music-based subjective evaluation expression dictionary 86 shown in FIG. 17C to delete impression expression words that are not registered in the same manner and collect similar impression expression words. The To generate a result list.
As shown in the figure, in the music thesaurus dictionary 85, nouns important in the music field are stored together with thesaurus information such as “rhythm = tempo” and “singer = singer”.
In the music-based subjective evaluation expression dictionary 86, impression expression words that are important in the music field, such as “uptempo” and “cool”, are stored together with similarity information.
[0116]
FIG. 17D shows an example of the result list 87 for song A and the result list 88 for song B. In this way, the CD purchase support apparatus first generates a result list for each song. Note that the sentence data of the song A shown in FIG. 17A is only an example, and the impression expression / noun phrase set includes only “up-tempo / rhythm” and “cool / singer”. Actually, text data including many impression expression / noun phrase sets is used.
[0117]
It can be seen from the result list 87 that the song A has an image such as “rhythm” is “up tempo” and “singer” is “cool”. Further, it can be seen from the result list 88 that the song B has an image such as “adult” in “musical tone”.
It can be said that the impression expression word / noun phrase set for each song output here is an image in which each song is generally leaned.
[0118]
Next, the CD purchase support apparatus merges all the text data and generates a result list for the merged data in order to provide an evaluation viewpoint when the user searches for music.
The evaluation viewpoint refers to the side of the object on which the image is held when an image is held on the object.
For example, there are “rhythm”, “singer”, “musical tone”, “atmosphere” and the like as evaluation viewpoints that have an image for a song.
In order to obtain such an evaluation viewpoint, as shown in FIG. 18A, the data input unit 2 merges (combines) all the collected sentence data of each song, and generates merge data.
[0119]
The CD purchase support apparatus extracts the impression expression / noun phrase set, confirms the noun phrase, confirms the impression expression for the merge data, and outputs the merge result list 91 shown in FIG. To do.
The noun phrase confirming unit 11 collects impression expression words using nouns as nodes from the impression expression word / noun phrase set, and counts the frequency of the noun phrases.
[0120]
The CD purchase support apparatus includes an evaluation viewpoint determination unit (not shown), and determines an evaluation viewpoint using the merge result list 91 output from the output unit 13.
It can be said that the higher the frequency of the noun phrases output in the merge result list 91, the more it can be used as an evaluation viewpoint when expressing an image related to music. Therefore, the evaluation viewpoint determination unit, for example, selects the noun phrases having the highest frequency ranking as the top four as the evaluation viewpoint. Shi To adopt.
By presenting these adopted evaluation viewpoints to the user, it is possible to select an evaluation viewpoint that is important when the user selects a CD.
[0121]
FIG. 19A is a diagram showing an example of an evaluation viewpoint selection screen 95 displayed on a display provided in the CD purchase support apparatus.
The evaluation viewpoint selection screen 95 includes an evaluation viewpoint selection area 96. In the evaluation viewpoint selection area 96, the evaluation viewpoint adopted by the evaluation viewpoint determination unit is displayed. In the example of the figure, “rhythm”, “singer”, “musical tone”, and “atmosphere” are displayed as evaluation viewpoints. These evaluation viewpoints can be selected by clicking with a mouse.
[0122]
FIG. 19B is a diagram showing an example of a result screen 97 displayed when the evaluation viewpoint “rhythm” is selected on the evaluation viewpoint selection screen 95.
Result screen 97 Displays an impression expression word linked to the evaluation viewpoint, and a song name having a set of the evaluation viewpoint (noun phrase) and the impression expression word.
For example, in the case of the evaluation viewpoint “rhythm”, “up-tempo (song)”, “slow (song)”, “easy to paste (song)”, and the like are displayed as impression expressions. Then, for example, a song A having the noun phrase “rhythm” and the impression expression word “uptempo” as an impression expression word / noun phrase set is displayed.
These impression expression words and song titles are those searched by the CD purchase support apparatus from the result lists 87 and 88 and other song result lists using the noun phrase “rhythm” as a key. Thus, the CD purchase support apparatus assigns an impression expression word for each evaluation viewpoint.
[0123]
When the evaluation viewpoint is determined, it can be said that each impression expression word related to the evaluation viewpoint is an evaluation image in the evaluation viewpoint.
Therefore, the CD purchase support apparatus presents each evaluation with respect to the evaluation viewpoint selected by the user who desires to purchase the CD, and outputs a song having the evaluation viewpoint and the evaluation as a candidate.
In this way, the user can search for a song that he / she feels.
The CD purchase support apparatus is considered to have great utility value when selecting a store's BGM (Back Ground Music) in addition to providing music search means to CD purchasers.
If the displayed song name is selected by clicking with the mouse, the touch of the song or the entire song may be viewed.
[0124]
(2) When supporting the total coordination of the party
When planning a party, for example, you want to unify the image in cooking, BGM, decorating flowers, and other settings, for example, you want to make sure that the image you have for the decorating flower matches the general image There are many requests for linking images and objects.
Here, as an example, a method for searching for a flower to decorate a party from an image will be described.
The main part of the system used in this example is the same as the image information search unit 3 (FIG. 1) of the first embodiment. Therefore, the same constituent elements as those in the first embodiment will be described using the same reference numerals. The hardware configuration is the same as that of the first embodiment (FIG. 9).
[0125]
First, at a party, information about each element necessary for setting is collected and stored in the storage device 58 or the like. Elements required for the setting include cooking, BGM, flowers, and lighting.
Here, in order to search for flowers to decorate at the party, first, image information (text data) relating to “flowers” is collected and stored.
Text data can be collected, for example, by searching the web using the name of a flower such as “lily, cosmos,...” As a key, and the retrieved text data is stored in a storage medium or the like for each type of flower. Keep it.
[0126]
The set extraction unit 10 extracts an impression expression word / noun phrase set from the sentence data acquired by the search unit 7.
However, when the impression object is a flower, the image is often expressed only by an impression expression word such as “dainty” or “grotesque”. Therefore, if there is no noun phrase that strongly relates to the impression expression word, only the impression expression word is extracted. In this case, the appearance frequency of impression expression words is recorded.
[0127]
FIG. 20A is a diagram showing an example of an initial expression list 98 of an impression expression word / noun phrase set extracted from sentence data on “sunflower” and (only an impression expression word when there is no related noun phrase). is there.
It can be seen that “bright / flower”, “strong / image”, “energetic”, “summerish / feel”, etc. were extracted from the text data on sunflower. Note that only “impression expression” is extracted from “Genki”. It can be seen that “bright” appears 20 times.
[0128]
noun phrase The confirmation unit 11 deletes an inappropriate noun phrase, aggregates impression expression words having noun phrases as nodes, and aggregates noun phrases using thesaurus information using a floral thesaurus dictionary related to flowers.
Then, the impression expression word confirmation unit 12 deletes impression expressions that are not appropriate in the flower field and aggregates similar impression expressions using the subjective evaluation expression dictionary. At this time, frequency information of impression expression words is also acquired. It can be said that the impression expression word with a higher frequency is more general as an image of the target flower.
Then, a threshold is set for the frequency, and an impression expression word having a frequency higher than the threshold is adopted as the image of the flower.
[0129]
FIG. 20B is a diagram showing an example of the result list 99 generated by the impression expression word confirmation unit 12. The impression expression words “bright” and “energetic” are registered as having a high degree of similarity in the subjective evaluation expression dictionary. Therefore, the impression expression word confirmation unit 12 collects “bright + energetic”. Further, the frequency after aggregation is obtained by adding the frequencies of “bright” and “energetic”. In this way, when the impression expression words are aggregated, the frequencies are also aggregated.
It can be seen from the result list 99 that sunflowers generally have a bright and energetic image.
As described above, for each flower such as a lily, cosmos, rose, chrysanthemum, tulip,..., An image in which the flower is generally leaned is acquired.
The acquired image is stored in the database by linking the image with the flower, and the corresponding flower can be searched when the image is input.
[0130]
FIG. 21 is a diagram showing an example of a search screen 105 for searching for a flower that matches the party atmosphere.
The query input column 106 is a column in which a query is input in a natural sentence as to what kind of party atmosphere the user wants to select or what image flower to select.
When the search button 107 is selected by clicking with a mouse operation or the like, the search is started and the result screen 108 of FIG. 21B is displayed.
In the example in the figure, as a result of inputting a query “feeling fine”, sunflowers were searched.
[0131]
Note that an image input as a query is also subjected to synonym control using the thesaurus dictionary unit 15 relating to flowers and the subjective evaluation expression dictionary.
In this case, in the system, the query “Energetic” is controlled to “bright”, and as a result, “sunflower” is output as a bright-feeling flower candidate.
[0132]
In the above description, the case of searching for flowers has been described as an example. However, there are various other things necessary for a party such as BGM, cooking, and so on. For these images as well, create a database of images that are generally held. For example, if you enter a query such as “bright feeling” in BGM, a song that generally holds a bright image is output. You can also.
[0133]
Alternatively, a database relating to each field such as “flower”, “BGM”, “cooking”,.
In this case, when the user inputs a field and an image such as “feeling bright in flowers”, the system searches a database relating to flowers and outputs the names of flowers that feel bright.
[0134]
Alternatively, for example, it is also possible to configure to output candidates such as flowers, BGM, dishes, etc. by inputting a single party image, such as “feeling bright”.
In addition, this embodiment is not limited to parties. It can be used when searching.
[0135]
(Fourth embodiment)
For example, assume that the content of an international tourist hotel includes the phrase “There is a Japanese garden and the food is delicious”.
From the fact that there is a “Japanese garden”, humans can presume that this hotel has “Miyabi” and “Japanese style”. However, the system according to the first to third embodiments cannot estimate an image unless there is an impression expression word such as “elegant Japanese garden”.
Therefore, the user can present an image recalled from a noun phrase without an impression expression word, or use it to create impression data. of It is difficult to make an estimate.
[0136]
Therefore, in this embodiment, a method for automatically causing the system to understand an image behind a noun such as “Japanese garden” will be described.
The system of the present embodiment can be configured with the main part using the image information search unit 3 (FIG. 1) of the first embodiment. Therefore, the same constituent elements as those in the first embodiment will be described using the same reference numerals. The hardware configuration is the same as that of the first embodiment (FIG. 9).
[0137]
First, in order to collect data to be analyzed, a large number of text data related to a predetermined theme (for example, accommodation facility) is collected by using a search engine on the Web, and these are merged.
[0138]
FIG. 22A is a diagram showing an example of collected sentence data. Thus, for example, even when it is desired to estimate impression data related to “OO hot spring”, text data related to the theme such as “ΔΔ hot spring” or “XX hot spring” is widely collected.
In the figure, one sentence data is shown for each accommodation facility. However, the present invention is not limited to this, and a plurality of sentence data can be collected for one accommodation facility.
Then, as shown in FIG. 22 (b), merged data is generated by merging the sentence data widely collected for the theme.
[0139]
The set extraction unit 10 extracts an impression expression word / noun phrase set from the merge data, and generates an initial set list.
FIG. 23 is a diagram showing an example of the initial set list extracted from the merge data. As shown in the figure, a set related to accommodation such as “beautiful / room”, “wide / open-air”, etc. has been extracted, and at this stage “good / weather” and “energetic” Sets that are not related to accommodation such as “/ child” have also been extracted.
[0140]
The noun phrase confirmation unit 11 aggregates the impression expression words using the initial set list as a node and generates an impression expression word aggregation list. Thereafter, noun phrases that are not registered in the accommodation facility thesaurus and impression expression words that are strongly related thereto are deleted. Furthermore, nouns such as synonyms are aggregated using thesaurus information, and a noun phrase aggregation list is generated. In addition, the frequency of noun phrases and impression expressions is acquired at the time of aggregation.
As a result, the nouns unnecessary for acquiring the image of the accommodation facility such as “weather” and “children” and the impression expression words strongly related thereto are deleted, and the noun phrase aggregation shown in FIG. You can get a list.
[0141]
The impression expression word confirmation unit 12 checks whether or not an impression expression word in the list of FIG. 24A exists in the impression expression word list using the accommodation facility type impression expression word dictionary. If it does not exist, delete it and make the frequency of the noun phrase correspond to it.
Furthermore, when a subjective evaluation expression dictionary is used as the impression expression word list, impression expression words having very close meanings can be aggregated.
[0142]
In FIG. 24A, “steep atmosphere” is not an impression expression word representing the atmosphere of the accommodation facility, and is not registered in the impression expression word list. Therefore, the impression expression word / noun phrase set “steep / atmosphere” is deleted from the list, and the result list shown in FIG. 24B is generated.
Then, the output unit 13 outputs the generated result list.
[0143]
As described above, it can be understood by a human that the “bad atmosphere” does not represent the atmosphere of the accommodation facility, but it is necessary to create a mechanism for the system to understand this.
Therefore, an appropriate expression for expressing the accommodation facility is listed in advance in the accommodation facility impression expression word dictionary, and an impression expression word that is not listed in this dictionary is determined not to be appropriate in the accommodation facility field. .
[0144]
For example, the impression expression word which is not important in the field of accommodation such as “rageous” is not listed, and the system does not list the extracted impression expression word “rageous” in the accommodation impression word dictionary So this impression expression word is appropriate Absent It can be judged.
That is, the system only needs to acquire the impression expression words listed in the accommodation expression system impression expression words from the extracted impression expression words.
[0145]
The result list in FIG. 24B is a result of collecting impression expression words such as key nouns and what kind of images the nouns have in the accommodation facility system.
In addition, this result list can be regarded as an image information value added to a picked-up noun.
For example, when a user imagines a “room” in an accommodation facility, it can be seen that problems such as “area”, “cleanliness”, and “calmness” are regarded as problems.
As described above, values such as “area”, “beautifulness”, and “calmness” are given as image information for the noun “room”.
Thus, it can be said that the result list indicates combinations of noun phrases and impression expression words that tend to be associated with the noun phrases.
[0146]
By using this result list, when a noun phrase not accompanied by an impression expression word is extracted, the impression expression word can be complemented.
For example, if there is an expression such as “There is a Japanese garden in this hot spring inn” without an impression expression word such as “A elegant Japanese garden”, , "Japanese garden" = "elegant image" can be understood from empirical or common sense knowledge, but generally computer systems cannot be understood, so no image can be acquired from this sentence.
[0147]
In such a case, referring to FIG. 24 (b), “gara”, which is the most frequently used impression expression word related to “Japanese garden”, is generally called “Japanese garden” by human beings. It turns out that the image is conceivable and this information can be automatically acquired by the computer system.
That is, “Japanese garden” is searched from the result list, and “Yana” that is combined with this is connected to “Japanese garden” extracted from the document data to complement the impression expression word. In the case where it is not desired to complement the image even when the “room” appears in a state where it is not associated with the impression expression word such as “there is a room”, the image is displayed from the state of FIG. A noun that complements the image and a noun that does not complement the image may be set in advance by setting a flag for the noun to be complemented.
Alternatively, only a noun for which an image is to be complemented can be selected to generate a new list, and this list can be used to complement the image.
[0148]
FIG. 25 is a diagram showing an example in the case of supplementing an image with text data.
The text data for acquiring an image of a certain accommodation facility includes the noun “Japanese garden” that does not have a strongly associated impression expression word.
The system of the present embodiment extracts an impression expression / noun phrase set (“bath / beautiful”) and a noun phrase “Japanese garden” without an impression expression from text data.
Then, the present system complements the image information for “Japanese garden” by adding “elegant” using the result list of FIG.
[0149]
In the fourth embodiment described above, the following effects can be obtained. First, it became possible for the system to understand the image. For example, human beings " You can understand images from nouns using common sense, such as “Japanese garden” = “Gorgeous”, “Chilli” = “Spicy”, “Express” = “Fast”, but computer systems generally capture images from noun phrases. I can't understand.
Therefore, in the present embodiment, by preparing a list in which noun phrases and impression expressions are linked, it is possible to cause the system to understand images that humans generally understand from noun phrases. As described above, in the fourth embodiment, it is possible to add image information to a noun phrase not accompanied by a strongly connected impression expression word in the sentence data.
In addition, although the above demonstrated using "accommodation facilities", it is not limited to this, Image information can be supplemented in various fields.
[0150]
As described above, in the first to fourth embodiments, the system can understand the image by creating a mechanism that links the noun phrase and the image. Furthermore, when the system understands the image, it has become possible to perform information processing on the image.
In the first to fourth embodiments, one embodiment of the present invention has been described. However, the present invention is not limited to the described embodiment, and the scope described in each claim. Various modifications can be made.
[0151]
【The invention's effect】
According to the present invention, it is possible to automatically acquire an expression representing an image generally held for a certain object.
[Brief description of the drawings]
FIG. 1 is a diagram illustrating an example of a logical configuration of an image information search device according to a first embodiment of the present invention.
FIG. 2 is a diagram showing an example of a query input screen.
[Fig. 3] Sentence data, It is the figure which showed one example of the initial set list.
FIG. 4 is a diagram schematically showing the structure of a thesaurus dictionary unit and a field-specific thesaurus dictionary.
FIG. 5 is a diagram for explaining aggregation of noun phrases performed by the noun phrase confirmation unit.
FIG. 6 is a diagram showing an example of a logical configuration of an impression expression word dictionary unit.
FIG. 7 is a diagram for explaining deletion of an impression expression word from a noun phrase aggregation list performed by an impression expression word confirmation unit.
FIG. 8 is a diagram showing an example of an output screen displaying the output result of the output unit.
FIG. 9 is a diagram showing an example of the configuration of the image information retrieval apparatus.
FIG. 10 is a flowchart for explaining the operation of the image information retrieval apparatus.
FIG. 11 is a diagram showing an example of a logical configuration of a subjective evaluation expression dictionary in a modification of the first embodiment of the present invention.
FIG. 12 is a diagram showing an impression expression word / noun phrase set, a noun phrase aggregation list, a result list, and the like according to the second embodiment of the present invention.
FIG. 13 is a diagram showing an example of a brand image screen.
FIG. 14 is a diagram showing the difference in image for each company in a pie chart.
FIG. 15 is a table in which images for the company brand are tabulated according to the user's age.
FIG. 16 is a diagram showing the change over time of the brand image.
FIG. 17 is a diagram illustrating sentence data, an impression expression word / noun phrase set, a music thesaurus dictionary, a music subjective evaluation expression dictionary, a result list, and the like according to the third embodiment of the present invention.
FIG. 18 is a diagram for explaining how to obtain an evaluation viewpoint.
FIG. 19 is a diagram showing an example of an evaluation viewpoint selection screen and a result screen.
FIG. 20 is a diagram for explaining acquisition of sunflower image information.
FIG. 21 is a diagram showing an example of a search screen and a result screen.
FIG. 22 is a diagram showing collected sentence data and merge data according to the fourth embodiment of the present invention.
FIG. 23 is a diagram illustrating an example of an impression expression word / noun phrase set extracted from merge data.
FIG. 24 is a diagram for explaining aggregation of impression expression word / noun phrase sets, and the like.
FIG. 25 is a diagram showing an example in the case where the image is complemented for the text data.
[Explanation of symbols]
1 Image information retrieval device
2 Data input part
3 Image information search part
5 Query input part
6 Field / Keyword Extraction Department
7 Search part
8 Search engine
10 set extraction unit
11 Noun phrase confirmation part
12 Impression expression confirmation part
13 Output section
15 Thesaurus dictionary
16 Impression expression word dictionary
21 Query input field
22 Search button
23 Query input screen
25 Text data
26 Initial set list
30 Car Thesaurus Dictionary
31 hot spring thesaurus dictionary
32 Beer Thesaurus
35 Impression expression word aggregation list
36 Noun phrase aggregation list
41 Vehicle impression expression dictionary
37 results list
45 Image information display column
46 Search field
47 output screen
48 Search result column
51 Control unit
52 ROM
53 CPU
54 RAM
55 Input device
56 Output device
57 Communication control unit
58 Storage device
59 Program storage
60 Data storage
61 Storage medium drive
62 I / O interface
63 Bus line
65 Impression expression word aggregation list
66 Noun phrase aggregation list
67 results list
71 Brand image screen
72 Brand Selection Area
73 Image information display area
74 Image information search field
75 Search button
85 Musical Thesaurus
86 Musical Subjective Evaluation Expression Dictionary
87 results list
88 results list
91 Merge result list
95 Evaluation viewpoint selection screen
96 Evaluation viewpoint selection area
97 result screen
98 Initial set list
99 results list
105 Search screen
106 Query input field
107 Search button
108 Results screen

Claims (8)

A search object acquisition means for acquiring an impression expression word as a keyword and a field for searching the impression expression word;
Sentence acquisition means for acquiring one or more sentences related to impression expression words as keywords acquired by the search target acquisition means ;
A plurality of impression expression words included in the acquired sentence, an extraction means for extracting the searched impression expression words and a noun phrase related to the impression expression words as a set;
Using a dictionary that associates noun phrases and fields, noun phrases specifying means for specifying those belonging to the acquired field among the noun phrases extracted in the set;
In the extracted set, output means for outputting an impression expression word that is set with the specified noun phrase;
An information processing apparatus comprising:   The information processing apparatus according to claim 1, wherein the output unit outputs the impression expression words in a sentence format using the impression expression words or a list format in which the impression expression words are listed. Impression expression word aggregation means for aggregating impression expression words set with the extracted noun phrases as nodes,
Rearrangement information creating means for creating rearrangement information for rearranging the aggregated impression expressions in a predetermined order;
Further comprising
The information processing apparatus according to claim 1, wherein the output unit determines an impression expression word to be output using the created rearrangement information. Comprising noun phrase aggregation means for aggregating noun phrases included in the set using thesaurus information;
4. When collecting noun phrases by the noun phrase collecting means, impression expression words that are combined with the aggregated noun phrases are also collected. Information processing device.   The said output means does not output what is not contained in the impression expression word dictionary which registered the impression expression word in the said field | area among the impression expression words contained in the said set. The information processing apparatus according to any one of claims 1 to 4. It further comprises a similarity acquisition means for acquiring a similarity between impression expression words included in the impression expression word dictionary,
The information processing apparatus according to claim 5, wherein impression expressions that are combined with the same noun phrase are aggregated using the similarity. The extraction means further comprises a function of extracting an impression expression word without a noun phrase involved,
When the extraction means extracts an impression expression word that is not accompanied by a noun phrase that is involved, the impression expression word is searched from a combination of a noun phrase and an impression expression word that are prepared in advance, and the searched impression The information according to any one of claims 1 to 6, wherein a noun phrase combined with an expression word is set as an impression expression word not accompanied by the noun phrase. Processing equipment. In a computer comprising search object acquisition means, sentence acquisition means, extraction means, noun phrase identification means, and output means,
In the search target acquisition means, and the impression expression word as a keyword, a retrieval target acquisition step of acquiring the field to search for the impression expression word,
In the sentence acquisition means, a sentence acquisition step of acquiring one or a plurality of sentences related to the impression expression word as a keyword acquired in the search target acquisition step ;
An extraction step of searching for a plurality of impression expression words included in the acquired sentence by the extraction means, and extracting the searched impression expression words and noun phrases related to the impression expression words as a set; ,
In the noun phrase specifying means, a noun phrase specifying step for specifying a noun phrase that belongs to the acquired field among the noun phrases extracted in the set using a dictionary that associates the noun phrase with the field,
In the output means, in the extracted set, an output step of outputting an impression expression word that is set with the specified noun phrase;
An information processing method comprising:

Images