JP3882414B2 - Document summarization apparatus, document summarization method, and recording medium - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a document summarization apparatus, a document summarization method, and a recording medium on which a document summarization program is recorded, which creates a summary of an input document and outputs the summary. In particular, the present invention summarizes by synthesizing selected binary relations. The present invention relates to a document summarizing apparatus, a document summarizing method, and a recording medium on which a document summarizing program is recorded.
[0002]
[Prior art]
When a document is searched by keyword search or the like, the search result includes many documents that do not meet the search purpose. Therefore, the searcher must grasp the contents of the document output as the search result and determine whether or not the document is a document suitable for the purpose. However, when a large amount of documents are extracted as search results, it is not easy to refer to all the search results and grasp the contents. Therefore, document summarization techniques have been proposed as a method for allowing readers to grasp the contents of retrieved documents in a short time. As such document summarization techniques, keyword enumeration, important sentence pickup, sentence shortening, etc. are common.
[0003]
The keyword enumeration method attempts to make the contents of a document roughly understood by enumerating important words such as frequent words in the document. As such a method, for example, “Cutting, et al.,“ Scatter / Gather: A Cluster-based Approach to Browsing Large Document Collections ”SIGIR-92, 1992” is known. In this method, a document is expressed as a collection of frequent keywords and a part of the article title. However, the keyword enumeration method does not show the relationship between the displayed frequent keywords, and therefore there is a problem that it is difficult to grasp the exact contents.
[0004]
In the method using important sentence pickup, an important sentence is determined based on a frequent word or sentence pattern, and a summary is generated by picking them up and connecting them. For example, Japanese Patent Application Laid-Open No. 02-093866 discloses a method for picking up a sentence including keywords that are registered in advance and that are semantically important. Japanese Patent Application Laid-Open No. 06-348606 discloses a method of scoring each word in a document and picking up a certain number of sentences having a high total score. According to such a method, since a summary in which important sentences are listed is output, specific contents can be obtained as compared with the case where only keywords are displayed. However, the method using the important sentence pickup has a problem that the selected sentence tends to be a long sentence and the contents cannot be grasped in a short time.
[0005]
As a method for sentence shortening, for example, in Japanese Patent Application Laid-Open No. 05-101120, a document abstract for shortening a sentence by deciding the type of dependency between what is omitted and what is not, and deleting a clause determined to be omitted. A generating device is disclosed. Japanese Patent Application Laid-Open No. 06-035961 discloses a document summarizing apparatus that deletes modifiers and shortens sentences. Furthermore, Japanese Patent Application Laid-Open No. 08-212228 discloses a summary sentence creating apparatus that creates a summary sentence by combining elements having high importance. Furthermore, Japanese Patent Laid-Open No. 06-259423 discloses an automatic summary sentence creation method in which an unnecessary part is deleted from an important paragraph after it has been selected. According to such a method, a document from which a less important part is deleted is displayed, so that an outline of the document can be quickly understood.
[0006]
[Problems to be solved by the invention]
However, the sentence shortening method has a problem that a large compression rate cannot be expected because it only deletes the modified part.
[0007]
In addition, since any method determines whether or not to delete only by the type of dependency, there is a problem that an important part may be excessively deleted. That is, in many cases, whether or not a word represents important information in the sentence cannot be determined merely by the dependency relationship, so that there is a possibility that important information is not displayed by these methods.
[0008]
The present invention has been made in view of these points, and an object thereof is to provide a document summarization apparatus that generates a summary that is simple and can sufficiently grasp the contents of a document.
[0009]
Another object of the present invention is to provide a document summarization method for generating a summary that is simple and can sufficiently grasp the contents of the document.
Furthermore, another object of the present invention is to provide a recording medium on which is recorded a program that causes a computer to generate a summary that is simple and can sufficiently grasp the contents of a document.
[0010]
[Means for Solving the Problems]
In the present invention, in order to solve the above problems, Summarized Analyzing the text and extracting the binary relation between words, and the binary relation extracted by the relation analyzing means Degree Relationship importance evaluation means to be given, Of the binary relations extracted by the relation analysis means, the relation importance evaluation means High importance Degree Granted Heavy Binary relations The Relationship selection means to select and complementary binary relations to be complemented Corresponding the complement rule for selecting and the priority of the complement rule Recorded relationship priority recording means; For each important binary relation selected by the relation selection means, Recorded in the relationship priority recording means Applying the complement rules according to the priority, until the complement binary relation to be complemented reaches a predetermined number Relationship complementing means for complementing the complementary binary relationship to the important binary relationship; Above Important binary relations and supplemented complementary binary relations Based on the text Summary Generation Phrase synthesizing means and , A document summarizing apparatus is provided.
[0011]
here , Seki The analysis method is Summarized Analyzing text to extract binary relations between words, relation importance evaluation means are important for binary relations extracted by relation analysis means Degree The relationship selection means Of the binary relations extracted by the relation analysis means, the relation importance evaluation means High importance Degree Granted Heavy Binary relations The Select and relationship priority recording means will be complemented complementary binary relation Corresponding the complement rule for selecting and the priority of the complement rule Record and supplement the relationship For each important binary relation selected by the relation selection means, Recorded in the relationship priority recording means Apply completion rules according to priority, until the complement binary relations to be complemented reach a predetermined number Complementary binary relations are supplemented to important binary relations, and phrase synthesis means are important binary relations and complemented complementary binary relations. Based on text Summary Generation To do.
[0012]
Also, Relational analysis means analyzes the text to be summarized to extract binary relations between words, relational importance evaluation means assigns importance to the binary relations extracted by the relational analysis means, and selects relations A means selects an important binary relation given high importance by the relation importance evaluation means from among the binary relations extracted by the relation analysis means, and a relation complementing means is selected by the relation selection means. Applying a complementary rule for selecting a complementary binary relation to be complemented for each of the important binary relations according to the priority associated with the complementary rule. Complementing the complementary binary relationship to the important binary relationship until a number is reached, and phrase synthesizing means generates a summary of the text based on the important binary relationship and the complemented complementary binary relationship; A document summarization method is provided.
[0013]
By doing so, it is possible to sufficiently complement the extracted binary relation, and as a result, it is possible to generate a concise summary having a sufficient amount of information.
[0014]
further, A computer that analyzes text to be summarized and extracts a binary relationship between words; a relationship importance evaluation unit that assigns importance to the binary relationship extracted by the relationship analysis unit; and the relationship analysis Among the binary relations extracted by the means, a relation selection means for selecting an important binary relation given a high importance by the relation importance evaluation means, and a complementary rule for selecting a complementary binary relation to be complemented And a relation priority recording means that records the priority of the complementing rule in association with each other, and the supplementary rule recorded in the relation priority recording means for each important binary relation selected by the relation selecting means. Applying according to the priority, relationship complementing means for complementing the complementary binary relation to the important binary relation until the complementary binary relation to be complemented reaches a predetermined number, the important binary relation and the complemented relation The phrase synthesizing means for generating a summary of the text based on the complementary binary relations, be functional computer-readable recording the document summary program characterized as A recording medium is provided.
[0015]
By doing so, it is possible to sufficiently complement the extracted binary relation, and as a result, it is possible to generate a concise summary having a sufficient amount of information.
[0016]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, embodiments of the present invention will be described with reference to the drawings.
FIG. 1 is a configuration diagram showing a configuration of a document summarizing apparatus 1 according to the present embodiment.
[0017]
The document summarization apparatus 1 includes a document input unit 1a for inputting a document, a text included in the document input from the document input unit 1a, and a relation analysis unit 1b for extracting a binary relation between words, a relationship analysis unit The relationship importance evaluation means 1c for assigning the importance score to the binary relation extracted by the above, the relation selection means 1d for selecting the binary relation given the high importance score as the important binary relation, and the complemented complementary binaries The relationship priority recording unit 1f that records the priority of the relationship that the relationship has, the relationship complementing unit 1e that complements the complementary binary relationship to the important binary relationship according to the priority recorded in the relationship priority recording unit 1f, and a summary Termination condition determining means 1g for determining whether or not a condition necessary for combining is satisfied, and phrase synthesizing means 1h for combining the summary from the important binary relation and the complemented complementary binary relation Thus it is constructed.
[0018]
FIG. 2 shows the contents of the relationship priority table 10 recorded by the relationship priority recording means 1f. The relationship priority table 10 includes part-of-speech 11 of words constituting a relational side and a reception side of a binary relation, a relation side / reception side 12 indicating the relation side and the reception side, a relation 14 connecting the relation side and the reception side, and The priority 13 is recorded. An integer greater than or equal to 0 is used for the expression of priority 13, and 0 means that complementation is essential, and the larger the number, the lower the complement priority.
[0019]
Next, the operation of the document summarizing apparatus 1 will be described using FIG. 1 and FIG.
First, a document to be processed is input to the document input unit 1a. The document input means 1a to which the document is input extracts only the text excluding figures and the like from the input document. Here, when the input document includes a tag such as an HTML document, the document input means 1a extracts only the text information from which the tag is removed. The text of the input document extracted in this way is sent to the relation analysis means 1b.
[0020]
The relation analysis unit 1b that has received the text of the input document analyzes the received text and extracts a binary relation between words. The level of analysis includes dependency analysis, syntax analysis, case analysis, semantic analysis, and the like. In this embodiment, dependency analysis for analyzing only the relationship between words is performed. In this embodiment, the binary relation is extracted from all sentences. However, an important part is selected in advance in arbitrary units such as sentences and paragraphs, and only the part is analyzed and the binary relation is extracted. It may be done. The binary relation extracted in this way is sent to the relation importance evaluation means 1c.
[0021]
The relationship importance evaluation means 1c that has received the binary relationship assigns an importance score to the binary relationship of the received word. Any method of scoring the importance score may be used. For example, the sum of the word importance by the tf * IDF product (G. Salton. “Automatic Text Processing.” Addison-Wesley (1989)) or the like may be used. A method of multiplying the relationship importance given for each type of relationship is used.
[0022]
The relationship selection means 1d selects the binary relation given the highest score from the binary relations given the importance score by the relation importance evaluation means 1c as the important binary relation that becomes the center of the phrase. To do. The important binary relation thus selected is sent to the relation complementing means 1e.
[0023]
The relationship complementing unit 1e that has received the important binary relationship refers to the relationship priority recording unit 1f, and complements the complementary binary relationship that needs to be complemented to the important binary relationship. As a specific procedure, first, a dependency side and a reception side having an important binary relation are extracted, and the part of speech is extracted. Next, the extracted dependency side, reception side, and their parts of speech are made to correspond to the relationship priority table 10, and a relationship 14 having a priority 13 corresponding to 0 is extracted. Then, a binary relation having the relation extracted in this way and the relevant side or the receiving side word of the important binary relation is selected from the binary relation extracted by the relation analysis means 1b and selected. The binary relation is complemented as an important binary relation as a complementary binary relation.
[0024]
Next, the relationship complementing unit 1e determines whether or not the number of complementary binary relationships corresponding to one important binary relationship that has been complemented in this way has reached a preset number. The number of settings here is automatically set by the system according to the purpose of using the summary, the user's preference, the type of core relationship, the part of speech on the side of interest, the part of speech on the side of the receiver, etc. It is good. Here, when the number of complemented binary relations that have been complemented does not reach the set number, the next highest priority relation is selected from the relation priority table 10, and the relation and the corresponding important binary relation are selected. A binary relation having a relational word or a receiving word is selected from the binary relations extracted by the relation analysis means 1b, and the selected binary relation is supplemented to the important binary relation as a complementary binary relation. . Then, until the number of complemented binary relations that have been complemented reaches the set number, relationships with higher priorities are sequentially selected, and similar complementation is performed. Then, when the number of complemented binary relations that have been complemented reaches the set number or the relation to complement is lost, the relation complementing means 1e ends the important binary relation and the complementary binary relation that complements it. It sends to the condition judgment means 1g.
[0025]
The termination condition determination unit 1g determines whether or not the selected important binary relation and the complementary binary relation that complements the selected important binary relation satisfy the termination condition necessary for generating the summary. Here, the termination condition can be the number of important binary relations that make up the summary, the number of relations that make up the summary, the number of characters in the summary, the ratio of the number of characters in the summary to the whole document, etc. May be. Here, if the termination condition is not satisfied, the processes after the relationship importance evaluation unit 1c are repeated until the termination condition is satisfied. When the processing after the relationship importance evaluation means 1c is repeated, the relationship importance evaluation means 1c drops a certain number of importance scores of the already selected important binary relation and the complementary binary relation complementing this, and the relation analysis is performed. An importance score is again assigned to the binary relation extracted by the means 1b. As a method of dropping the importance score, for example, a method of multiplying the importance of a word having an already selected important binary relation and a complementary binary relation complemented by a constant R (0 <R <1). Is used. In this way, it is possible to prevent re-selecting the already selected binary relation by dropping a certain number of importance scores of the already selected important binary relation and the complementary binary relation complementing it. it can. When the important binary relation selected in this way and the complementary binary relation that complements the important binary relation satisfy the end condition necessary for generating the summary, the end condition determining means 1g selects all the important binary relations selected. The term relation and the complementary binary relation for complementing them are sent to the phrase synthesizing means 1h.
[0026]
The phrase synthesizing unit 1h that has received the important binary relation and the complementary binary relation satisfying the termination condition synthesizes them to generate a summary.
Next, the operation of the document summarization apparatus 1 will be described using a flowchart. FIG. 3 is a flowchart showing an operation procedure in which the document summarizing apparatus 1 generates a summary.
[0027]
[S1] The document input unit 1a extracts only text from the input document and sends the text to the relationship analysis unit 1b. The relation analysis unit 1b that has received the text performs dependency analysis on the received text in units of documents.
[0028]
[S2] The relational analysis means 1b extracts a binary relation between words from the dependency analysis result. The extracted binary relation is sent to the relation importance evaluation means 1c.
[S3] The relationship importance evaluation unit 1c that receives the binary relationship gives an importance score to the binary relationship extracted by the relationship analysis unit 1b.
[0029]
[S4] The relationship selecting unit 1d selects the binary relationship having the highest score as the important binary relationship, and sends the selected important binary relationship to the relationship complementing unit 1e.
[S5] The relationship complementing unit 1e refers to the relationship priority recording unit 1f and supplements the essential complementary binary relationship with respect to the important binary relationship.
[0030]
[S6] The relationship complementing unit 1e determines whether or not the number of complemented binary relations complemented is equal to or greater than a preset number. If the number is greater than or equal to the specified number, the selected important binary relation and the complemented complementary binary relation are sent to the termination condition determining means 1g, and the process proceeds to step S9. If the specified number has not been reached, the process proceeds to step S7.
[0031]
[S7] The relationship complementing unit 1e determines whether or not the relationship having the next highest priority remains in the relationship priority table 10. If it remains, the process proceeds to step S8. If not, the process proceeds to step S9.
[0032]
[S8] The relation complementing means 1e is a binary relation extracted by the relation analyzing means 1b that has a high-priority relation and a relevant or binary word of the important binary relation. The selected binary relation is complemented to the important binary relation as a complementary binary relation.
[0033]
[S9] The termination condition determination unit 1g determines whether the termination condition is satisfied. When it is satisfied, all the selected important binary relations and the complementary binary relation that complements them are sent to the phrase synthesizing means 1h, and the process proceeds to step S10. When not satisfy | filling, it progresses to step S3.
[0034]
[S10] The important binary relation and the complementary binary relation satisfying the termination condition are synthesized to generate a summary.
Next, the operation of the document summarization apparatus 1 will be described using an actual document example.
[0035]
In the following, the summary generation operation will be explained for the document example, “Yasujiro Ozu took this reviewer from Cathay Hotel with his disliked bird's-eye view.”
First, a document is input to the document input means 1a, “Yasujiro Ozu took this reviewer from Cathay Hotel with his disliked bird's-eye view”. The document input means 1a to which this document is input extracts only the text from the input document. Since this document example is composed only of text, a document having the same content as the input document is extracted as text. The text extracted in this way is sent to the relationship analysis means 1b.
[0036]
The relation analysis unit 1b that receives the text performs dependency analysis of the received text. FIG. 4A shows the result of dependency analysis performed by the relationship analysis means 1b. The relation analysis unit 1b that has finished the dependency analysis of the text next extracts a binary relation from the dependency analysis result. FIG. 4B shows the extracted binary relationship. In FIG. 4B, the binary relation is expressed as ([Relationship between words], a relational word, a receiving word). Here, the relationship between words is expressed by the surface case of the case relationship, the particle that connects the words, the combination of the part of speech on the side of the dependency and the side of reception. Depending on the level of analysis, the case relationship may be expressed by a deep case such as “actor”, or a more detailed relationship may be used. The binary relation extracted in this way is sent to the relation importance evaluation means 1c.
[0037]
The relationship importance evaluation means 1c that has received the binary relationship gives an importance score to the received binary relationship. Here, scoring is performed by a method such as tf * IDF product.
[0038]
The relationship selecting unit 1d selects the binary relationship having the highest score as the important binary relationship, and sends the selected important binary relationship to the relationship complementing unit 1e. In this example, assume that the binary relation (taken by Yasujiro Ozu) was selected as the important binary relation.
[0039]
The relationship complementing unit 1e that received the important binary relationship refers to the relationship priority recording unit 1f, and complements the complementary binary relationship with respect to the important binary relationship. The relationship complementing means 1e first complements the complementary binary relationship essential for the important binary relationship. As a procedure for complementing the complementary binary relation, first, the part-of-speech of the dependency side and the receiving side in the important binary relation is extracted. Here, the important binomial relation ([taken by Yasujiro Ozu]) has been selected, so the clerk is the noun “Yasujiro Ozu” and the receiver is the verb “taken”. Here, referring to the relationship priority table 10 shown in FIG. 2 for the dependency side, the relationship priority table 10 does not have the relationship 14 when the part of speech on the dependency side is a noun. Next, referring to the relationship priority table 10 for the receiver, a plurality of relationships 14 in the case where the part of speech of the receiver is a verb are shown. Of those relationships 14, [GA], [O], [HA], and [MO] are selected as the relationship 14 with the priority 13 that is essential to be complemented. Then, the relational analysis means is used to select the relation 14 ([GA], [O], [HA], [MO]) and the binary relation having the receiver “taken” with the priority 13 selected in this way. When selecting from the binary relation shown in FIG. 4 (b), where 1b is the extracted binary relation, the relation [[]] (taken by [[]] reviews) is extracted as a complementary binary relation. The Rukoto.
[0040]
When the completion of the essential complementary binary relation is completed, the relation complementing means 1e then reaches the preset number of complementary binary relations corresponding to the important binary relation thus complemented. Judge whether or not. Here, assuming that the set number is set to 2, the number of complementary binomial relationships that have been complemented so far is 1, and the set number 2 has not been reached. Therefore, the relationship complementing unit 1e performs complementation of the complementary binary relationship having the next highest priority. Of the relationships 14 in the case where the part of speech on the receiving side is a verb, the relationship 14 with the next highest priority 13 after 0 is [i] with a priority 13 of 1. However, the binary relation having [Ni] and the receiver “taken” does not exist in the binary relation shown in FIG. Next, the relationship 14 with the higher priority 13 is [to] with the priority 13 of 2. The binary relationship corresponding to this is also the binary relationship shown in FIG. Does not exist. Then, when [De], which is the relationship 14 with the next highest priority 13, is selected, the binary relationship (taken with a bird's-eye view) with this [De] and the receiver “taken” is shown in FIG. It exists in the binary relation shown in b), and this binary relation (taken overlooking [[]]) is complemented as a complementary binary relation.
[0041]
When ([taken a bird's-eye view)] is complemented, the relationship complementing means 1e again sets the number of complementary binary relations corresponding to the important binary relations thus complemented in a preset number. It is determined whether or not a certain 2 is reached. Here, since the complemented complementary binary relation has reached 2, the process of the relation complementing means 1e is terminated, and the important binary relation and the complementary binary relation for complementing it are sent to the termination condition determining means 1g.
[0042]
The termination condition determination means 1g that has received the important binary relation and the complementary binary relation that complements the important binary relation, the important binary relation selected to generate the summary and the complementary binary relation that complements the important binary relation are predetermined. It is determined whether the termination condition is satisfied. If it is determined that the end condition is not satisfied, the steps from step S3 to step S8 in the flowchart shown in FIG. 3 are repeated until the end condition is satisfied. At this time, the importance score of the already selected important binary relation and complementary binary relation is dropped. In this example, if it is assumed that the termination condition is not satisfied, the words “Ozu Yasujiro”, “Reviewer”, “Overview”, “Take” have already been selected. The importance level is reduced by, for example, multiplying by a constant R (0 <R <1). When the end condition is satisfied, the end condition determining unit 1g sends all the selected important binary relationships and the complementary binary relationship that complements them to the phrase synthesizing unit 1h.
[0043]
The phrase synthesizing unit 1h that has received the important binary relation and the complementary binary relation satisfying the termination condition synthesizes the sent important binary relation and the complementary binary relation, and generates a summary. In this example, (as taken by Yasujiro Ozu) was selected as the important binary relation, and (as a supplementary binary was taken) and (taken from a bird's-eye view) as complementary binary relations to complement it. Is selected, and a summary that combines them, “Yasujiro Ozu took a review with a bird's-eye view,” is generated.
[0044]
In the above-described example, only one phrase is synthesized, so there is only one phrase, but actually, a plurality of phrases are synthesized according to the end condition. These are displayed in an array corresponding to the order of appearance of the input document.
[0045]
Also, in the above example, there is no relationship that should be complemented by the words on the dependency side. You may make it determine the side to complement by the kind of binary relationship. Moreover, you may make it always give priority to one side, for example, a receiving side.
[0046]
Thus, in this embodiment, the relational analysis means 1b extracts a binary relation from the text included in the input document, and the relational importance evaluation means 1c gives an importance score to the extracted binary relation, The relational selection means 1d selects the binary relation having the highest importance score as the important binary relation, and the relation complementing means 1e selects the important binary relation according to the priority of the relation recorded in the relation priority recording means 1f. Complementary binary relations are supplemented, and the phrase composition means 1h generates the summary by synthesizing the important binary relations and the complementary binary relations. Even if the number is not large or the number is small, complementation is performed according to the priority order recorded in the relationship priority recording unit 1f until the complement binary relation to be complemented reaches the set number. As a result, it is possible to collect binary relations sufficient to generate a summary, and to generate a summary that is concise and has a sufficient amount of information for grasping the document contents. .
[0047]
In this form, it is divided into mandatory relationships and other relationships with priorities, and the essential relationships are always complemented. However, priorities are assigned to all relationships, and a certain number from the highest priority is assigned. You may make it complement.
[0048]
In this embodiment, the example in which the priority of the relationship is determined based on the part of speech and whether it is the dependency side or the reception side has been described. However, the priority may be roughly or more finely classified. For example, when the important binary relation has [From], the priority of [To] corresponding to this may be increased.
[0049]
Furthermore, in this embodiment, an example has been described in which different priorities are given to all non-essential relationships. For example, the same priority is given to a paired relationship such as [From] and [To], and one of them is complemented. If done, the other may be complemented.
[0050]
Further, in this embodiment, every time one important binary relation is selected, the complementary binary relation is complemented. However, all the binary relations are selected in advance by the relation selection means 1d, and then each complementary binary relation is selected. It is also possible to complement the term relation.
[0051]
The above processing functions can be realized by a computer. In this case, the processing contents of the functions that the document summarizing apparatus should have are described in a program recorded on a computer-readable recording medium. Then, by executing this program on a computer, the above processing is realized by the computer. Examples of the computer-readable recording medium include a magnetic recording device and a semiconductor memory. When distributing to the market, store the program in a portable recording medium such as a CD-ROM (Compact Disk Read Only Memory) or floppy disk, or store it in a computer storage device connected via a network. In addition, it can be transferred to another computer through the network. When executed by a computer, the program is stored in a hard disk device or the like in the computer, loaded into the main memory, and executed.
[0052]
【The invention's effect】
As described above, in the document summarization apparatus of the present invention, the relation analysis means Summarized Extract binary relations from the text and use the relation importance assessment tool to extract the binary relations Degree And by relationship selection means Of the extracted binary relations important Degree High Heavy Binary relations The Selected and recorded in the relationship priority recording means by the relationship complementing means Apply completion rules according to priority , Complementary binary relations are supplemented to important binary relations, and important binary relations and complementary binary relations are obtained by means of phrase synthesis. On the basis of the Since the summary is generated, even if the document does not have the required modifiers that should be supplemented to the important binary relation, or the number of supplementary binary relations to be complemented is the set number Will be supplemented according to the priority order recorded in the relationship priority recording means, and as a result, sufficient binary relations can be collected to generate a summary, which is concise and understands the document contents It is possible to generate a summary with a sufficient amount of information for.
[0053]
In the document summarization method of the present invention, Summarized Extract binary relations from text, important for extracted binary relations Degree Grant, Of the extracted binary relations important Degree High Heavy Binary relations The Selected and recorded in the priority recording means Apply completion rules according to priority , Complementary binary relations to important binary relations, important binary relations and complementary binary relations On the basis of the Since the summary is generated, even if the document does not have the required modifiers that should be supplemented to the important binary relation, or the number of supplementary binary relations to be complemented is the set number Will be supplemented according to the priority order recorded in the relationship priority recording means, and as a result, sufficient binary relations can be collected to generate a summary, which is concise and understands the document contents It is possible to generate a summary with a sufficient amount of information for.
[0054]
Furthermore, in the recording medium of the present invention, Summarized Extract binary relations from text, important for extracted binary relations Degree Grant, Of the extracted binary relations important Degree High Heavy Binary relations The Selected and recorded in the relationship priority recording means Apply completion rules according to priority , Complementary binary relations to important binary relations, important binary relations and complementary binary relations On the basis of the Since the program that causes the computer to perform the function of generating the summary is recorded, by invoking the program recorded on the recording medium on the computer, the essential modifiers that the document should complement the important binary relations are added. Even if it does not have, or even if the number is small, it will be complemented according to the priority order recorded in the relationship priority recording means until the complement binary relation to be complemented reaches the set number, result As a result, it is possible to collect binary relations sufficient to generate a summary, and it is possible to generate a summary that is simple and has a sufficient amount of information for grasping document contents.
[Brief description of the drawings]
FIG. 1 is a configuration diagram showing a configuration of a document summarizing apparatus.
FIG. 2 is a diagram showing the contents of a relationship priority table recorded by a relationship priority recording means.
FIG. 3 is a flowchart showing an operation procedure for generating a summary by the document summarization apparatus.
FIG. 4 is a diagram showing a specific example of dependency analysis and binomial relationship extraction performed by a relationship analysis unit.
[Explanation of symbols]
1 Document summary device
1a Document input means
1b Relational analysis means
1c Relationship importance evaluation means
1d Relation selection means
1e Relationship complementing means
1f Relation priority recording means
1g End condition judging means
1h phrase synthesis means
10 relationship priority table

Claims (4)

A relationship analysis means for analyzing the text to be summarized and extracting a binary relationship between words;
A relation importance evaluation means for assigning importance to the binary relation extracted by the relation analysis means;
Among the binary relation that the relationship analyzing means has extracted, the relationship selecting means for selecting the relationship importance estimation means by a high degree of importance important binomial granted relationship,
A relationship priority recording means for recording a complementary rule for selecting a complementary binary relationship to be complemented and a priority of the complementary rule ;
For each of the important binary relations selected by the relation selection means, the supplementary rules recorded in the relation priority recording means are applied according to the priority, and the complemented binary relations to be complemented become a predetermined number. Relationship complementing means for complementing the complementary binary relationship to the important binary relationship until reaching
And clauses synthesizing means for generating a summary of the text based on the important binary relations and complemented the complementary binary relations,
A document summarization apparatus comprising: And further comprising an end condition determining means for determining whether the important binary relation complemented by the relation complementing means and the complemented binary relation supplemented have reached a predetermined amount of information,
The phrase synthesizing means is based on the important binomial relation that has been complemented by the time when the end condition judging means determines that the predetermined amount of information has been reached and the complemented binary relation that has been complemented. Generate a text summary,
The document summarization apparatus according to claim 1, wherein: The relationship analysis means analyzes the text to be summarized to extract the binary relationship between words,
A relationship importance degree evaluation unit assigns importance to the binomial relationship extracted by the relationship analysis unit;
The relationship selection means selects an important binary relation given high importance by the relation importance evaluation means from among the binary relations extracted by the relation analysis means,
A relationship complementing unit applies a complementary rule for selecting a complementary binary relationship to be complemented for each of the important binary relationships selected by the relationship selecting unit according to the priority associated with the complementary rule, Supplementing the complementary binary relationship to the important binary relationship until the complementary binary relationship to be complemented reaches a predetermined number;
Phrase synthesizing means generates a summary of the text based on the important binary relation and the complemented complementary binary relation;
A document summarization method. Computer
A relationship analysis means that analyzes the text to be summarized and extracts the binary relationship between words;
A relation importance evaluation means for assigning importance to the binary relation extracted by the relation analysis means;
Of the binary relations extracted by the relation analysis means, a relation selection means for selecting an important binary relation given a high importance by the relation importance evaluation means,
A relationship priority recording means for recording a complementary rule for selecting a complementary binary relation to be complemented and a priority of the complementary rule;
For each of the important binary relations selected by the relation selection means, the supplementary rules recorded in the relation priority recording means are applied according to the priority, and the complemented binary relations to be complemented become a predetermined number. Relationship complementing means for complementing the complementary binary relationship to the important binary relationship until reaching
Phrase synthesizing means for generating a summary of the text based on the important binary relation and the complemented complementary binary relation;
A computer-readable recording medium on which a document summarizing program is recorded.

Images