JP2885489B2 - Document content search device - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a document content search device that extracts necessary information from stored information by inputting a question sentence and provides it to a user.

[Conventional technology]

In this type of conventional document content search device, a device for searching a database stored in advance organized for each item as a search target or a text for searching by calculating the similarity between a registered sentence and an input sentence A type database device (for example, Japanese Patent Application No. 1-1111626) is known.

[Problems to be solved by the invention]

However, in the former device, a database that is organized and stored in advance for each item is targeted for retrieval.
Information stored as text could not be searched. In addition, in order to search the database, processing is performed to convert it to a database search command using a world model or the like that describes field knowledge. To perform such conversion correctly, a complicated data structure is required. We needed to build a knowledge base. With the latter device, sentences with high semantic similarity can be searched.
What products have been released? ], It was not possible to give an accurate answer to the question sentence format.

An object of the present invention is to provide information from textual information.
An object of the present invention is to provide a document content search device that can provide an accurate answer to an input question sentence and can be realized without building a knowledge base itself having a complicated data structure.

[Means for solving the problem]

For each information sentence in the input document file, an information sentence pattern is determined, and the document content table creator extracts the item contents corresponding to the item attributes described in the information sentence pattern, for each information sentence pattern. Creates and accumulates a document content table that describes the relationship between the information sentence, item attributes, and item contents corresponding to, and extracts the word information and phrase information of the input question sentence. A question sentence pattern is determined with reference to the inter-pattern relation table, an item content corresponding to the item attribute described in the question sentence pattern is extracted, and an information sentence pattern corresponding to the determined question sentence pattern is obtained. The information sentence that belongs to the information sentence pattern that has the same item attribute as the item sentence of the question sentence pattern is determined with reference to the document content table. In information statement, by extracting the item content of the item attribute corresponding to the answer attribute corresponding to the determined question pattern, based on the extracted contents, and configured to display the answer content.

[Action]

An information sentence that matches the item content of the determined question sentence pattern is obtained, and then the item content of the item attribute that matches the answer item attribute corresponding to the determined question sentence is extracted from the information sentence. Therefore, an accurate answer can be made to the input question sentence from the information described in the sentence.

〔Example〕

FIG. 1 is a block diagram showing an embodiment of the present invention. Hereinafter, the operation of the embodiment of the present invention will be described with reference to FIG. In the figure, 1 is a word dictionary unit, 2 is a morphological analysis unit, 3 is an information sentence analysis unit, 4 is an information sentence pattern dictionary unit, 5 is a pattern relation table unit, 6 is an information sentence pattern determination unit, and 7 is item contents. Extraction unit, 8 is a document content table creation unit, 9 is a document content table storage unit, 10 is a question sentence analysis unit, 11 is a question sentence pattern determination unit, 12
Denotes an answer extraction unit, and 13 denotes an answer display unit.

FIG. 2 is a diagram showing an example of a document file. Second
In the figure, a sentence number is assigned to each sentence when it is read by the information sentence analysis unit 3, and is used as an ID for each information in the following processing. Here, each sentence in the document file is called an information sentence, and the information sentence with the sentence number i is described as an information sentence i.

 First, the information sentence analysis processing will be described.

The information sentence analysis unit 3 reads information from the document file and sends it to the morphological analysis unit 2 to obtain word information and analysis information.

The morphological analysis unit 2 refers to the word dictionary unit 1 and extracts notation of a word constituting the analyzed sentence, word information including the part of speech and the semantic attribute of the word, and phrase information.

FIG. 3 is a diagram showing an example of the relationship between the notation described in the word dictionary unit 1 and the semantic attributes. For example, the same semantic attribute [sale] is assigned to "sale" and "release". This semantic attribute is used for determining an information sentence pattern and a question sentence pattern, which will be described later.

FIG. 4 is a diagram showing an example of the processing result of the information sentence analyzing unit 3 for the information sentence "NTT released the Howdy cordless phone from May 1." The information sentence is divided into words, a phrase number is given as phrase information, and a part of speech,
The semantic attributes are extracted. Words with the same phrase number indicate that they are in the same phrase. For example, "NTT" and "GA"
Are in the same phrase, and have the parts of speech “proper noun”,
"Particles" are extracted. As for “release”, “verb stem” is extracted as a part of speech, and “selling” is extracted as a semantic attribute. The “howdy cordless phone” is a word that is not registered in the word dictionary unit 1 and is an unknown word that is regarded as one word depending on the context, the type of character, and the like, and information on the part of speech and the meaning attribute cannot be obtained.

 Next, the information sentence pattern determination processing will be described.

The information sentence pattern determination unit 6 determines an information sentence pattern by referring to the information sentence pattern dictionary unit 4 based on the word information obtained by the information sentence analysis unit 3.

FIG. 5 is a diagram showing an example of information sentence patterns stored in the information sentence pattern dictionary unit 4. The information sentence pattern dictionary unit 4 stores an information sentence pattern represented by a combination of phrase patterns composed of columns of item attributes, words, and semantic attributes. For example, in FIG. 5, two information sentence patterns A
1, examples of A2 are shown. Here, Di in the sentence pattern is
An item attribute is shown, and a character string following the item attribute indicates a word whose part of speech is "particle", and one of the phrase patterns is represented by the item attribute and the word. Data enclosed by [] indicates a semantic attribute, which also represents one phrase pattern. For example, in the case of information sentence pattern A1, item attribute D1
A sentence pattern is represented by a combination of a phrase pattern 'D1' containing 'and a phrase pattern' D2 'containing an item attribute D2 and a semantic attribute [sales].

Various methods can be applied to the pattern determination method, such as a determination based on whether a specified condition is satisfied or a determination based on whether the similarity is calculated to be equal to or greater than a certain threshold. Here, the set of semantic attributes included in the information sentence T is
An example will be described in which, when there is an information sentence pattern P that includes a set of semantic attributes of the information sentence pattern, the information sentence pattern of the information sentence T is determined to be P.

 For example, the set Mt of the semantic attributes of the information sentence 1 is Mt = {[hour], [selling]}.

On the other hand, a set M1 and M2 of semantic attributes of information sentence patterns A1 and A2
Is M1 = {[selling]} M2 = {[partnership]}.

Since M1 is included in Mt, the information sentence pattern of information sentence 1 is
A1 is determined.

 Next, the document content table creation processing will be described.

The document content table creator 8 first extracts the item contents from each information sentence as follows.

Using the information sentence as an extracted sentence, and using the information sentence pattern determined by the information sentence pattern determination unit 6 as a reference sentence pattern, the item content extraction unit 7 is started to extract the item contents.

The item content extraction unit 7 compares the word information extracted from the extracted sentence and the phrase information with the phrase pattern in the reference sentence pattern, thereby extracting the item content corresponding to the item attribute in the reference sentence pattern. Extract from the extracted sentence.

Here, a method of extracting item contents based on particles in a phrase pattern will be described. In the case of the information sentence 1, since the information sentence pattern is A1, the item content extraction unit 7 is started using the information sentence 1 as an extracted sentence and the information sentence pattern A1 as a reference sentence pattern. For example, in the item content extraction unit 7,
The process of extracting the item contents for the item attribute D1 is performed as follows. The item attribute D1 belongs to a phrase pattern "D1 ga" containing the particle "ga". In the information sentence 1, the phrase containing the particle "GA" is "NTT ga". Therefore, replacing D1 with "NTT" results in the same phrase. In this way,
A character string that matches a phrase by replacing a certain character string with an item attribute is defined as the item content of this item attribute. In the case of this example, the item content of the item attribute D1 in the information sentence 1 is “NTT”. In this way, the item contents in each information sentence are extracted. If a plurality of identical particles exist, the determination is made in consideration of the relationship between the phrase patterns.

Next, the document content table creator 8 creates, for each information sentence pattern, a document content table that describes the information sentences belonging to each information sentence pattern and the item content of each item attribute for each information sentence pattern.

FIG. 6 is a diagram showing an example of a document content table.
For example, information sentence 1 and information sentence 2 belong to information sentence pattern A1, and the item contents of item attributes D1 and D2 of information sentence 1 are each “N”.
TT "and" howdy cordless phone ", and the item contents of the item attributes D1 and D2 of the information sentence 2 are" NTT "and" answering machine ", respectively. Although "howdy cordless phone" is an unknown word, if the phrase information is correct, the item contents can be correctly extracted as described above.

The document content table created in this way is stored in the document content table storage unit 9.

Next, the document content search processing will be described. FIG. 8 is a diagram showing an example of a document content search process, which will be described below with reference to FIG.

 First, the question sentence analysis process will be described.

The question sentence analyzer 10 sends the input question sentence to the morphological analyzer 2 and obtains word information and phrase information. For example, if the question Q "What products are sold by NTT?"
Phrase information and word information are extracted like "question sentence analysis result" in the figure. At this time, assuming that the set of semantic attributes of the question text is I, I = {[selling], [product]}.

 Next, a process of determining a question sentence pattern will be described.

The question sentence pattern determination unit 11 refers to the word information obtained by the question sentence analysis unit 10 and the question sentence patterns in the inter-pattern relation table stored in the inter-pattern relation table unit 5, and determines which question sentence It is determined whether it belongs to the question sentence pattern. Next, item attributes included in the question sentence pattern,
An information sentence pattern and an answer item attribute corresponding to the question sentence pattern are extracted.

FIG. 7 is a diagram showing an example of an inter-pattern relation table stored in the inter-pattern relation table section 5. FIG. 7 shows an example of three question sentence patterns Q1, Q2, and Q3.
The inter-pattern relation table section 5 is defined as a question sentence pattern by a combination of phrase patterns composed of columns of item attributes, words, and semantic attributes. Further, for each question sentence pattern, an ID of an information sentence pattern to be referred to and The answer item attribute is defined. If the question sentence belongs to which question sentence pattern from the pattern relation table, it indicates which information sentence pattern belongs to which information sentence pattern and which item attribute (answer item attribute) should be referred to. I have. For example, the question sentence pattern Q1 is a phrase pattern 'D
1 'is defined as a set of and semantic attributes [sell] and [product],
A1 is defined as an information sentence pattern to be referenced, and D2 is defined as an answer item attribute.

Here, a set of semantic attributes of each question sentence pattern Q1, Q2, Q3
Assuming N1, N2, and N3, N1 = {[Sell], [Product]} N2 = {[Affiliation], [Company]} N3 = {[Sell], [Company]}.

In the question sentence pattern determining unit 11, a question sentence pattern is determined by a process similar to the process of the information sentence pattern determining unit 6.
Since the set I of semantic attributes of the question sentence Q includes the set N1 of semantic attributes of the question sentence pattern Q1, the question sentence pattern of the question sentence Q is
Q1 is determined. Further, A1 is obtained as an information sentence pattern to be referenced, and D2 is obtained as an answer item attribute.

 Next, the answer extraction processing will be described.

The answer extraction unit 12 first extracts the item contents of the question sentence as follows.

The item content extraction unit 7 is started by using the question sentence as the sentence to be extracted and the question sentence pattern determined by the question sentence pattern determination unit 11 as the reference sentence pattern, and extracts the item contents.

The extraction method is the same as in the case of the information sentence described above.
For example, in the case of the question sentence Q, the extracted part is the question sentence Q,
The reference sentence pattern is a question sentence pattern Q1. At this time, "NTT" is extracted as the content of the item attribute D1 (see "item content extraction result" in FIG. 8).

Next, in each item attribute in the question sentence pattern, an information sentence in which the item content of the question sentence matches the item content of the information sentence belonging to the information sentence pattern to be referred to is determined.

For example, in the case of the question sentence Q, the information sentence pattern to be referred to is A1. By referring to the document content table, the information sentences belonging to A1 are found to be the information sentences 1 and 2. The item content of the item attribute D1 of each information sentence is “NTT” in information sentence 1 and “ATT” in information sentence 2. Therefore, for the question sentence Q, the information sentence 1 that matches the item content “NTT” of the item attribute D1 of the question sentence Q is determined.

Next, the item content of the item attribute that matches the answer item attribute is extracted from the item content of the determined information sentence. For example, in the case of the question sentence Q, since the answer item attribute is D2, the item content “howdy cordless phone” of the item attribute D2 of the determined information sentence 1 is extracted.

Finally, the answer display section 13 displays the answer contents based on the extracted item contents. For example, in the case of question Q,
“Howdy Cordless Phone” is displayed.

〔The invention's effect〕

As described above, according to the present invention, it is possible to easily extract necessary information from accumulated information by inputting a question sentence with information described in a sentence as a search target. In a conventional device, a search cannot be performed without constructing a database in which methods to be searched are arranged in advance for each item. However, the device according to the present invention analyzes sentences in a document file and executes an information sentence pattern. Is determined, the item contents of the item attributes defined by the information sentence pattern are extracted, a document contents table is created, and the document contents table is searched for. And can be. Therefore, the use of the apparatus according to the present invention has an effect that an information service capable of retrieving document contents can be provided in a short time even for a large amount of information sent daily, such as newspaper articles.

In addition, since the conventional device searches using the search command,
When searching by inputting a question sentence, it is converted into a search command using a knowledge base having a complicated data structure such as a world model. However, in the apparatus according to the present invention, a pattern relation table is used. By describing the relationship between a question sentence pattern and an information sentence pattern, a question sentence can be input and searched, so that a device that can search the contents of a document without requiring a great deal of effort for building a knowledge base. Can be easily constructed.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 is a diagram showing an example of a document file, FIG. 3 is a diagram showing an example of the relationship between the notation of a word dictionary and semantic attributes, FIG. Is a diagram showing an example of a processing result of the information sentence analyzing unit, FIG. 5 is a diagram showing an example of an information sentence pattern, FIG. 6 is a diagram showing an example of a document content table, and FIG. Figure showing an example of a table,
FIG. 8 is a diagram showing an example of a document content search process. 1. Word dictionary unit, 2. Morphological analysis unit, 3. Information sentence analysis unit, 4.
Information sentence pattern dictionary part, 5… Pattern relation table part, 6
… Information sentence pattern judgment unit, 7… item content extraction unit, 8… document content table creation unit, 9… document content table storage unit, 10… question sentence analysis unit, 11… question sentence pattern judgment unit, 12… answer extraction unit , 13… An answer display part.

Claims (1)

(57) [Claims] 1. A morphological analysis unit for extracting word information and phrase information comprising a notation of a word constituting a parsed sentence, a part of speech and a semantic attribute of the word with reference to a word dictionary, and information in a document file. An information sentence analysis unit that reads out a sentence and sends it to the morphological analysis unit to obtain word information; and an information sentence pattern that stores an information sentence pattern represented by a combination of a clause pattern consisting of a column of item attributes, words, and semantic attributes A dictionary part, a question sentence pattern represented by a combination of phrase patterns composed of columns of item attributes, words, and semantic attributes; an answer item attribute representing the information sentence pattern corresponding to the question sentence pattern; A pattern relation table unit that stores a pattern relation table that defines the information, word information obtained by the information sentence analysis unit, and an information sentence pattern dictionary that is stored in the information sentence pattern dictionary unit. An information sentence pattern determining unit that determines which information sentence pattern each information sentence belongs to with reference to the information sentence pattern, and based on the word information and phrase information extracted from the extracted sentence,
An item content extraction unit for extracting an item content corresponding to an item attribute in the reference sentence pattern from the extracted sentence by performing a matching with a phrase pattern in the reference sentence pattern; Is used as an extracted sentence, and the information sentence pattern determined by the information sentence pattern determination unit for the information sentence is used as a reference sentence pattern. For each pattern, a document content table creating unit for creating a document content table describing information sentences belonging to the information sentence pattern and item contents of each item attribute for each information sentence; A question sentence analysis unit for sending word information and phrase information for the question sentence, a word information obtained by the question sentence analysis unit, and an inter-pattern relation table stored in the inter-pattern relation table unit. The question text pattern in the table is referred to to determine which question text pattern the question text belongs to, and the item attributes included in the question text pattern, the information text pattern corresponding to the question text pattern, and the answer A question sentence pattern determining unit for obtaining an item attribute, the question sentence as an extracted sentence, and the question content pattern determined for the question sentence by the question sentence pattern determining unit as a reference sentence pattern; Activate the section and extract the item contents corresponding to each item attribute in the question sentence pattern, and in each item attribute in the question sentence pattern, correspond to the information sentence pattern obtained by the question sentence pattern determination section. An information sentence whose item content in the document content table of the information sentence matches the item content extracted from the question sentence is determined, and the answer item attribute and the answer item attribute are determined from the item content of the information sentence. A document content search device comprising: an answer extraction unit for extracting item contents of a matching item attribute; and an answer display unit for displaying answer contents based on the item contents extracted by the answer extraction unit. .