JP2011257791A - Relational information extracting device, its method and its program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a device capable of extracting relational information between a plurality of characteristic expressions in all texts, and to provide its method and its program.SOLUTION: A device which extracts relational information of a plurality of input characteristic expressions comprises: an analysis processing unit 10 for performing a morphological analysis of an input text and analyzing dependency of clauses composing the input text, when the text including each characteristic expression is input; and a relational information extraction processing unit 20 for extracting a clause including an independent word as a relational information candidate from a sentence different from a sentence including each characteristic expression in a plurality of sentences composing the input text, storing a noun clause which is extracted in order from a beginning of a sentence in the input text in a storage area mapped with a particle associating with the noun clause until extracting all of each characteristic expression and relational information candidate, acquiring storage information indicating a storage status of the noun clause in the storage area when extracting all of each characteristic expression and relational information candidate, and extracting the relational information based on an analysis result and the storage information.

Description

  The present invention relates to a technique for extracting information related to a plurality of specific expressions, which plays an important role in a summarization system or the like that summarizes input text.

  Conventionally, as a technique for extracting information related to a plurality of specific expressions, a dependency structure of input text, and information indicating whether or not words included in the input text are easily used as a relationship between specific expressions, There is known one that extracts relational information by machine learning using a feature as a feature.

  For example, if the text “Japan's Prime Minister Hatoyama will visit Korea next month” and a set of unique expressions “Hatoyama” and “Japan” are entered, the phrase “Japan” The information “Prime” is based on the dependency structure that is related to the phrase “Prime” and the information that the word “Prime” is easy to use as an indication of the relationship between the person and the organization. It is extracted as information indicating the relationship of the pair.

  In addition, when the relation information extraction technique described above is used, when a set of specific expressions “Hatoyama” and “Korea” is input, the information “visit” is extracted as the relation information.

Toru Hirano and two others, “Identification of relations between proper expressions using relative nounness”, Proc. Of the 15th Annual Conference of the Association for Natural Language Processing, March 2009, D5-5

  However, in the technique described in Non-Patent Document 1, it is possible to extract relation information when each unique expression and relation information exist in the same sentence, but each unique expression and relation information exist in different sentences. In this case, since the dependency relationship between the clause including each unique expression and the clause including the relationship information is unknown, the optimum relationship information cannot be extracted.

  It should be noted that the inventor found that each of the unique expressions and relational information out of a total of 11,236 <proprietary expression 1, proper expression 2, relational information> included in 8,150 randomly extracted texts. When the number of triples of <specific expression 1, specific expression 2, relation information> existing in the same sentence was investigated, each specific expression and relation information existed in the same sentence <specific expression 1, specific expression 2, relation The number of triples of information> was 9,178. That is, in the prior art, the relationship information can be extracted only from about 82% of all triplets, and it is difficult to extract the optimum relationship information from any text. It was.

  The present invention has been made in view of the above problems, and an object of the present invention is to provide an apparatus, a method, and a program that can extract relation information between a plurality of specific expressions in any text. .

  In order to achieve the above object, the related information extracting apparatus of the present invention is an apparatus for extracting information related to a plurality of input specific expressions, and when a text including each of the specific expressions is input, An analysis processing unit that analyzes the morphological analysis of the text and analyzes the dependency of the clauses that make up the input text, and when the analysis result by the analysis processing unit is acquired, a sentence that includes each unique expression among the plurality of sentences that make up the input text After extracting phrases containing independent words from other different sentences as related information candidates, the noun phrases extracted in order from the beginning of the input text until each of the proper expressions and related information candidates are extracted Stored in the storage area associated with the particle associated with, and storage information indicating the storage status of noun phrases in the storage area when all of the specific expressions and candidate relational information are extracted Acquired by, and a relationship information extraction processing section for extracting relevant information from the related information candidates based on the analysis results and stores the information.

  Further, the related information extraction method of the present invention is a method for extracting information related to a plurality of input specific expressions using a computer in order to achieve the object, wherein the computer Input text, morphological analysis of the input text and analysis of the dependency of the clauses that make up the input text, in other sentences that are different from the sentences that contain each specific expression among the sentences that make up the input text After extracting phrases containing independent words as related information candidates, associate noun phrases extracted in order from the beginning of the input text with the particle associated with the noun phrase until all the specific expressions and related information candidates are extracted. Storage information indicating the storage state of the noun phrase in the storage area is obtained when all of the specific expressions and relation information candidates are extracted. And extracting the relationship information from the related information candidates based on the analysis results and stores the information.

  Furthermore, the program of the present invention is for causing a computer to function as each means of the related information extracting apparatus.

  Furthermore, the program of the present invention is for causing a computer to execute each process of the related information extraction method.

  As a result, a phrase including an independent word is extracted as a related information candidate from another sentence different from the sentence including each specific expression among a plurality of sentences constituting the input text, and each specific expression and When all of the related information candidates are extracted, the related information is extracted from the related information candidates based on the storage information indicating the storage state of the noun phrase in the storage area and the analysis result, so that the noun stored in the storage area Based on the storage state of clauses, it becomes possible to extract relation information from relation information candidates existing in sentences different from each unique expression.

  According to the related information extracting apparatus, method and program of the present invention, based on the storage state of noun phrases stored in the storage area, the related information is extracted from the related information candidates existing in a sentence different from each unique expression. So, for example, if you enter a text that says, “Next, a confrontation of interest. A dream-like plan by President Suzuki and Director Takahashi.” “Suzuki” and “Takahashi” It is possible to extract the relation information “confrontation” existing in a sentence different from the sentence including the set of proper expressions. Therefore, it is possible to extract relation information between a plurality of unique expressions in any text.

1 is a configuration diagram of a relationship information extraction device according to a first embodiment of the present invention. Relationship information extraction process flow chart Diagram showing the summary of analysis results by the dependency analysis unit The figure which shows the outline | summary of the processing result by a specific expression matching part The figure which shows an example of the data structure of SRL information The figure which shows the outline of the processing result by central topic information extraction section The figure which shows an example of the data structure of the SRL information in the 2nd Embodiment of this invention The figure which shows the outline | summary of the process result by the central topic information extraction part in the 2nd Embodiment of this invention. The block diagram of the related information extraction apparatus in the 3rd Embodiment of this invention Flow chart of related information extraction processing in the third embodiment of the present invention The figure which shows the outline of the processing result by the relation estimation information acquisition part

  1 to 6 show a first embodiment of the present invention. FIG. 1 is a configuration diagram of a relational information extraction apparatus according to the present embodiment, FIG. 2 is a flowchart of relational information extraction processing, and FIG. FIG. 4 is a diagram showing an overview of the analysis result by the analysis unit, FIG. 4 is a diagram showing an overview of the processing result by the specific expression association unit, FIG. 5 is a diagram showing an example of the data structure of SRL information, and FIG. It is a figure which shows the outline | summary of the process result by.

  The outline of the related information extracting apparatus and method according to the present invention will be described below with reference to the drawings.

  The related information extracting apparatus of the present invention comprises a computer device mainly composed of a well-known CPU, and can be connected to display means such as a monitor, input means such as a keyboard, storage means such as a hard disk and memory, and an external network. A device (not shown) is provided. In addition, the relationship information extraction apparatus of this embodiment includes an analysis processing unit 10, a relationship information extraction processing unit 20, and a model storage unit 30.

  As shown in FIG. 1, the analysis processing unit 10 includes a morphological analysis unit 11, a dependency analysis unit 12, and a specific expression association unit 13, and a combination of specific expressions composed of two specific expressions and text serve as input means. When input is used, the input text is subjected to morphological analysis and the dependency relation of clauses constituting the input text is analyzed. In the present embodiment, a set of specific expressions composed of two specific expressions “Suzuki” and “Takahashi” is input.

  When acquiring the input text (step S1-1 in FIG. 2), the morpheme analysis unit 11 divides the input text into words by performing a well-known morphological analysis process on the input text, and gives a part of speech to each divided word. And output (step S1-2 in FIG. 2). For example, if the text “Next is a confrontation of interest. A dream-like plan by President Suzuki and Director Takahashi” is entered, the result of processing by the morphological analysis unit 11 is “Next Ni (conjunctive) /, (reading) / attention (noun) / no (participant) / confrontation (noun) /. (Phrase) ”,“ that (combined noun) / Suzuki (noun) / president (noun) / and ( Particle) / Takahashi (noun) / director (noun) / and (particle) / i (verb) / u (verb suffix) / great (noun) / na (determination) / rich (noun) / by (particle) /, (Reading mark) / dream (noun) / no (participant) / you (auxiliary noun) / na (determination) / planning (noun) /.

  The dependency analysis unit 12 divides the text into phrases by performing a well-known dependency analysis process on the input text that has been obtained from the morpheme analysis unit 11 and has been subjected to morpheme analysis. The dependency relationship is analyzed and output (step S1-3 in FIG. 2). In this case, when the illustrated input text is analyzed by the dependency analysis unit 12, information representing a dependency structure (dependency tree) as shown in FIG. 3 is output as an analysis result. Here, the phrase “next /,” and the phrase “attention / no” are related to the phrase “confrontation /.”, And when these dependency relationships are implemented as data, For example, “(confrontation. (Next) (attention)”), and these dependency relationships include dependency type information defined in a well-known dependency analysis technique (normal "D" representing a dependency relationship, "P" representing a parallel dependency relationship, "A" representing a similar dependency relationship, or "Q" representing a self-relational relationship).

  The specific expression association unit 13 is for performing a specific expression association process using the input set of specific expressions and the analysis result of the dependency analysis unit 12. More specifically, when the specific expression association unit 13 acquires a set of specific expressions input by using the input unit and the input text analyzed by the dependency analysis unit 12 (step S1-4 in FIG. 2). Then, a specific expression corresponding to each input specific expression is extracted from the input text, and a specific expression identifier representing the specific expression is assigned to the extracted specific expression (step S1-5 in FIG. 2). For example, when a specific expression “Suzuki” is input, “Suzuki” in the input text is expressed as “<PSN> Suzuki </ PSN>”. Here, “PSN” is a unique expression identifier representing a person name. In this embodiment, eight types of unique expression identifiers (“PSN” representing a person name and “PSN” representing an organization name defined in a well-known unique expression extraction technique). ORG ”,“ LOC ”representing place name,“ ART ”representing artifact name,“ MNY ”representing amount,“ PNT ”representing percentage,“ TIM ”representing time, and“ DAT ”representing date) ing.

  In the present embodiment, a set of specific expressions is expressed as “Suzuki: Takahashi”. In this case, the proper expression ("Suzuki") that appears first in the input text in the set of specific expressions is displayed on the left side of ":" as the forward specific expression, and the specific expression ("Takahashi") that appears later is the backward specific expression. It is shown on the right side of “:”. In addition, the processing result of the specific expression association unit 13 in this embodiment is shown in FIG.

  Next, an outline of the relationship information extraction processing unit 20 will be described. The relationship information extraction processing unit 20 includes a relationship information candidate extraction unit 21, a central topic information extraction unit 22, a model selection unit 23, a classifier 24, and a relationship information extraction unit 25, and is acquired from the analysis processing unit 10. Based on the analysis result, the relation information of the set of specific expressions is extracted.

  When the relationship information candidate extraction unit 21 acquires the analysis result from the analysis processing unit 10 from the specific expression association unit 13, the relation information candidate extraction unit 21 is independent from other sentences different from the sentence including each specific expression among the plurality of sentences constituting the input text. A phrase including a word is extracted as a related information candidate (step S1-6 in FIG. 2). For example, in the present embodiment, when the analysis result shown in FIG. 4 is input, two phrases “attention” and “confrontation” are extracted as related information candidates. In the phrase “next”, the “next” part-of-speech is a conjunction, and is not an independent word. Therefore, the phrase “next” is not extracted as a related information candidate.

  When the central topic information extraction unit 22 acquires the analysis result by the analysis processing unit 10 from the specific expression association unit 13 and acquires the related information candidates from the related information candidate extraction unit 21, all of the specific expressions and the related information candidates are obtained. Until noun phrases are extracted in order from the beginning of the input text until they are extracted, they are stored in the storage area associated with the particles associated with the noun phrases. Further, the central topic information extraction unit 22 acquires storage information indicating the storage state of the noun phrase in the storage area when all the unique expressions and the related information candidates are extracted (step S1-7 in FIG. 2). When a plurality of related information candidates are extracted in step S1-6, stored information is acquired for each related information candidate.

  Here, in the present embodiment, the storage information is acquired based on SRL (Sale Referent List) information used for processing for specifying a subject omitted in Japanese text or the like. SRL information is used to identify a word that is the center of a topic among words that have already appeared in the input text, and the identified word is highly related to a word included in a subsequent sentence. Is determined. For details of SRL information and processing using it, see Shigeko Nariyama: Grammar for ellipsis resolution in Japanese, In Proceedings of the 9th International conference on Theoretical and Methodological Issues in Machine Translation, Kyoto Japan, 135-145 (2002). Please refer to.

  The storage means (not shown) for storing the SRL information includes a plurality of particles (in this embodiment, “ha”, “ga”, “ni”, “wo”, and “other” representing other particles). Each time a noun phrase is extracted in order from the beginning of the input text, each extracted noun phrase is stored in a LIFO stack format in the storage area associated with the particle associated with the noun phrase. Remember. When a noun phrase is newly stored in the storage area, the storage order of the noun phrase in the storage area is stored at the highest (first) position.

  Specific processing contents of the central topic information extraction unit 22 will be described. When the central topic information extraction unit 22 acquires the analysis result shown in FIG. 4 and the two related information candidates “attention” and “confrontation.”, A set of specific expressions “Suzuki: Takahashi” and “confrontation. Noun phrases are extracted from the beginning of the input text until all of the related information candidates “” are extracted. First, when the central topic information extraction unit 22 reads the noun phrase “attention” from the input text, as shown in FIG. 5, there is no storage area corresponding to the particle “no” associated with the noun phrase in the storage means. Therefore, the noun phrase “attention” is stored as the first record in the storage area corresponding to the particle “other”.

  Next, when the central topic information extraction unit 22 reads the noun phrase “confrontation”, as shown in FIG. 5, “confrontation” is recorded as the first record in the storage area corresponding to the particle “ha” associated with the noun phrase. Is memorized. Here, in the input text, “ha” does not accompany “confrontation”, but when the sentence of the input text is expressed in a verbal stop, it is considered that the particle “ha” representing the proposal is accompanied. Therefore, the noun phrase “confrontation” is stored in the storage area corresponding to the particle “ha”.

  Next, when the central topic information extraction unit 22 reads the noun phrase “President Suzuki” from the input text, a storage area corresponding to the particle “to” associated with the noun phrase exists in the storage unit as shown in FIG. For this reason, the noun phrase “attention” is stored as the second record in the storage area corresponding to the particle “other”. In this case, in the storage area corresponding to the particle “other”, the noun phrase “President Suzuki” is the highest-level record.

  Next, when the central topic information extraction unit 22 reads out the noun phrase “Director Takahashi”, the combination of the specific expression “Suzuki: Takahashi” and the related information candidate “confrontation” are extracted. Is determined. Further, based on the SRL information at this time, the central topic information extraction unit 22 acquires tree structure data representing the contextual structure of each unique expression and relation information candidate as storage information. Specifically, referring to FIG. 6, the central topic information extraction unit 22 reads the last base element of the SRL information shown in FIG. "Is placed in the root node 100.

  Then, the central topic information extraction unit 22 arranges the storage status of another unique expression “Suzuki” and related information candidates as child nodes in the SRL information. For example, for another specific expression “Suzuki”, a node 101 representing another specific expression is arranged as a child node of the root node 100. Further, a node 102 indicating which particle in the SRL information corresponds to which particle in the SRL information and a node 103 indicating a storage order in the storage area are arranged as child nodes of the node 101, respectively. Is done. Furthermore, regarding the relationship information candidate “confrontation”, the node 104 indicating that it is a relationship information candidate is arranged as a child node of the root node 100. In addition, a node 105 that represents the notation of the related information candidate, a node 106 that represents the part of speech of the related information candidate, a node 107 that represents in which storage particle the related information candidate is stored in the SRL information, and The nodes 108 representing the storage order in the storage area are arranged as child nodes of the node 104, respectively. In FIG. 6, the notation of each specific expression is represented by a specific expression class.

  In this way, the storage information about the related information candidates of the specific expression group “Suzuki: Takahashi” and “Confrontation” is acquired. Further, the storage information is acquired by performing the same processing as above for the relationship information candidate “attention” and the specific expression group “Suzuki: Takahashi”.

  When the model selection unit 23 acquires the processing result of the central topic information extraction unit 22, the model selection unit 23 classifies the set of proper expressions based on the specific expression identifier given by the specific expression association unit 13, and uses a classifier 24 described later. The type of model to be extracted is selected (step S1-8 in FIG. 2). For example, when “Suzuki: Takahashi” is input as a set of proper expressions, the model selection unit 23 classifies the set of specific expressions into a type of “person name: person name”, and the type of the set of classified proper expressions Is output.

  When the classifier 24 acquires the processing result of the central topic information extraction unit 22 and the processing result of the model selection unit 23, a plurality of models are obtained based on the types of the unique expressions selected by the model selection unit 23. A model is extracted from the stored model storage unit 30. And the classifier 24 discriminate | determines whether each relationship information candidate turns into the relationship information of the set of a specific expression using the extracted model (step S1-9 of FIG. 2).

  Here, the model uses the analysis processing unit 10 and the result including the result determined in advance for the relationship information corresponding to the predetermined specific expression set and the text including each specific expression corresponding to the predetermined specific expression set, and It is generated in advance by performing well-known machine learning using information extracted in advance from the relationship information extraction process 20. In addition, the discrimination result for a predetermined set of specific expressions is made in advance based on human judgment. Note that each model may be configured according to the type of set of unique expressions, such as “person name: place name”, “person name: name”, or the like, or may be configured without distinguishing the types.

  In this case, since the storage information acquired by the central topic information extraction unit 22 is used for the determination by the classifier 24, it is determined whether or not each relationship information candidate is related information of a set of unique expressions. The determination can be made based on the storage state of the noun phrases stored in the storage area. In the present embodiment, the relationship information candidate “confrontation” among the relationship information candidates is determined to be the relationship information of the specific expression set “Suzuki: Takahashi”.

  Note that the classifier 24 may be configured to output a numerical value indicating the degree of possible relation information in addition to the determination result of whether or not the relation information is obtained. As machine learning, a well-known machine can be used. However, it is desirable to use a machine that can learn by directly inputting data of a tree structure or a graph structure.

  The relationship information extraction unit 25 extracts a relationship information candidate determined by the classifier 24 as relationship information of a set of specific expressions from each relationship information candidate. Further, the relationship information extraction unit 25 extracts the words from the beginning of the phrase to the main word among the extracted relationship information candidates as relationship information and outputs it to the display means (step S1-10 in FIG. 2). In addition, when the classifier 24 is configured to output a numerical value indicating the degree that can be related information, the related information extraction unit 25 extracts a related information candidate having the largest numerical value as related information. Alternatively, relationship information candidates whose numerical values are larger than a predetermined threshold value may be extracted as relationship information.

  In this way, the text “Next, the confrontation to be noticed. That dream plan by the great rich man, President Suzuki and Director Takahashi.” And the specific expression pair “Suzuki: Takahashi” were entered. Then, the relationship information “confrontation” is extracted.

  As described above, in the above embodiment, the relation information can be extracted from the relation information candidates existing in a sentence different from each unique expression based on the storage state of the noun phrase stored in the storage area. Next, a confrontation of interest, a sentence that includes a pair of proper expressions "Suzuki" and "Takahashi" when the text "A dreamy plan by the great rich of President Suzuki and Director Takahashi" is entered. It is possible to extract the relationship information “confrontation” that exists in a sentence different from. Therefore, it is possible to extract relation information between a plurality of unique expressions in any text.

  Further, when all of the specific expressions and the related information candidates are extracted, the particles in the storage area in which the noun phrases corresponding to the related information candidates are stored are acquired as stored information. Since the particle associated with the noun phrase can be used as a feature, the extraction accuracy of optimum relation information can be improved.

  Furthermore, when all of the specific expressions and the related information candidates are extracted, the storage order in the storage area of the noun phrase corresponding to the related information candidate is acquired as the stored information, so that in the related information determination process, Since the storage order of noun phrases can be used as a feature, the extraction accuracy of optimum relation information can be improved.

  Furthermore, when all of the specific expressions and related information candidates are extracted, the part of speech of the related information candidate is acquired as storage information, so that the part of speech of the noun phrase is used as a feature when the related information is determined. Therefore, it is possible to improve the extraction accuracy of the optimum relationship information.

  The relationship information extraction device, method and program thereof according to the second embodiment of the present invention will be described below. The difference between the present embodiment and the first embodiment is that when the relationship information extraction processing unit 20 extracts all the unique expressions and relationship information candidates, among the noun phrases stored in the storage area, The point is that information representing the storage state of a noun phrase having a dependency relationship with any one of the specific expressions or the phrase corresponding to the relation information candidate is acquired as storage information. Since other configurations and operations are the same as those of the above-described embodiments, only differences will be described here.

  Since only the noun phrase is stored in the SRL information described above, when the related information candidate is a verb phrase, the related information candidate is not stored in the SRL information. Therefore, there is a possibility that optimal relation information cannot be obtained because storage information including relation information candidates cannot be obtained.

  Therefore, the central topic information extraction unit 22 according to the present embodiment, when all of the specific expressions and relation information candidates are extracted, any one of the specific expressions or the relation among the noun phrases stored in the storage area. Information indicating the storage state of a noun phrase having a dependency relationship with the phrase corresponding to the information candidate is acquired as storage information. Below, the specific processing content of the central topic information extraction part 22 is demonstrated.

  The central topic information extraction unit 22 obtains the analysis result by the analysis processing unit 10 (in this embodiment, a set of specific expressions “X: Y” is input) from the analysis processing unit 13 and “XX”. ”Is acquired from the relationship information candidate extraction unit 21, the noun phrase from the beginning of the input text is extracted until all of the unique information set“ X: Y ”and the relationship information candidate“ XX ”are extracted. Is extracted and stored in the storage area. Here, the relation information candidate “XX” is a verb clause related to the noun phrase “□□”, and the specific expression “Y” among the specific expression pairs and the related information candidates appears at the end of the input text. The SRL information when the specific expression “Y” is extracted is shown in FIG.

  Next, the central topic information extraction unit 22 acquires, as storage information, tree structure data representing the contextual structure of each unique expression and related information candidate based on the SRL information of FIG. Specifically, referring to FIG. 8, the central topic information extracting unit 22 reads out the base element of the SRL information shown in FIG. "Is placed in the root node 100.

  Then, the central topic information extraction unit 22 arranges the storage status of another unique expression “Suzuki” in the SRL information as a child node. For another specific expression “X”, a node 101 representing another specific expression is arranged as a child node of the root node 100. Further, a node 102 indicating which particle in the SRL information corresponds to which particle in the SRL information and a node 103 indicating a storage order in the storage area are arranged as child nodes of the node 101, respectively. Is done.

  In addition, since the related information candidate “XX” is a verb clause, it is not stored in the SRL information. However, the central topic information extraction unit 22 determines “□□”, which is a related clause of the related information candidate “XX”. Is used as a child node. That is, the central topic information extraction unit 22 arranges the node 110 indicating which particle in the storage area corresponding to the particle in the SRL information the dependency clause “□□” is stored as a child node of the root node 100. Further, the node 111 representing the relationship information candidate is arranged as a child node 111 of the node 110. Furthermore, a node 112 representing the notation of the relationship information candidate, a node 113 representing the part of speech of the relationship information candidate, a node 114 representing the dependency relationship of the relationship information candidate to the clause “□□”, and the clause “□□ in the storage area. And the node 115 representing the storage order of “” are arranged as child nodes of the node 111, respectively.

  In this way, the storage information about the related information candidates “x: y” and “xxx” of the specific expression is acquired. Then, the acquired storage information is transmitted to the model selection unit 23. The processing contents of the model selection unit 23, the classifier 24, and the relationship information extraction unit 25 are the same as those in the first embodiment. In this case, “XX” is extracted as the relationship information of the unique expression group “X: Y”.

  In the above embodiment, when all of the specific expressions and relation information candidates are extracted, the noun phrase stored in the storage area is modified with any one of the specific expressions or the phrase corresponding to the relation information candidate. Since the information indicating the storage state of the related noun phrase is acquired as the storage information, for example, even when the relation information candidate is a verb phrase, the storage information including the relation information candidate can be acquired. Accordingly, since it is possible to acquire the optimum relation information based on the stored information, the extraction accuracy of the relation information can be further improved.

  Hereinafter, a related information extracting apparatus, a method thereof, and a program according to the third embodiment of the present invention will be described. This embodiment is different from the first and second embodiments described above in that when the relation information extraction processing unit 20 acquires the analysis result by the analysis processing unit 10, each of the plurality of sentences constituting the input text is unique. After extracting a phrase including an independent word from another sentence different from a sentence including an expression as a related information candidate, relation estimation information indicating the degree to which the relation information candidate is estimated to be relation information and the storage information It is obtained for each extracted relationship information candidate, and the relationship information is extracted from the relationship information candidate based on the analysis result, the relationship estimation information and the stored information. Since other configurations and operations are the same as those of the above-described embodiments, only differences will be described here.

  As shown in FIG. 9, the relationship information extraction apparatus according to the present embodiment includes an analysis processing unit 10, a relationship information extraction processing unit 20, a model storage unit 30, and a relationship estimation information storage unit 40. .

  In addition, the related information extraction processing unit 20 includes a related information candidate extracting unit 21, a central topic information extracting unit 22, a model selecting unit 23, a classifier 24, a related information extracting unit 25, and a related estimation information acquiring unit 26. It consists of and.

  Next, the operation of the relationship information extraction processing unit 20 will be described with reference to the flowchart of FIG. Here, the processing contents of steps S2-1 to S2-7 and the processing contents of steps S2-9 to S2-11 in FIG. 10 are the same as the processing contents of steps S1-1 to S1-7 and step S1 in FIG. It is the same as each processing content of −8 to step S1-10.

  When the relationship estimation information acquisition unit 26 acquires the processing result of the central topic information extraction unit 22, the relationship estimation information acquisition unit 26 acquires, from the relationship estimation information storage unit 40, relationship estimation information indicating the degree to which the relationship information candidate is estimated to be the relationship information. (Step S2-8 in FIG. 10). The relationship estimation information storage unit 30 stores in advance a plurality of relationship estimation information corresponding to each of a plurality of morphemes, for example, “attention = 0.2”, “confrontation = 1”, and each relationship estimation. The information is created by a person or a calculation method described later. For example, when the relationship estimation information acquisition unit 26 acquires the storage information illustrated in FIG. 6, among the relationship estimation information stored in the relationship estimation information storage unit 30, the relationship corresponding to the relationship information candidate “confrontation.” The estimation information is searched, and the corresponding relation estimation information “1” is extracted. Then, the relationship estimation information acquisition unit 26 adds the extracted relationship estimation information as a child node 120 of the candidate node 104 in the stored information as shown in FIG. Note that the relationship estimation information acquisition unit 26 performs the same processing for other relationship information candidates (“attention”) extracted by the relationship information candidate extraction unit 21.

  As for the technique for calculating the estimated degree of relational information from a large corpus, Tanaka et al., “Classification of syntactic categories of nouns based on dispersion of semantic categories”, Transactions of Information Processing Society of Japan, vol. 40, no. 9, pp. 3387-3396, September 1999.

  As described above, in the above-described embodiment, when the analysis result by the analysis processing unit is acquired, the phrase including the independent word from the sentence other than the sentence including each unique expression among the plurality of sentences constituting the input text is related information. After extracting as a candidate, relation estimation information indicating the degree to which the relation information candidate is estimated to be relation information and the storage information are acquired for each extracted relation information candidate, and an analysis result, relation estimation information, and Since the relationship information is extracted from the relationship information candidates based on the stored information, the degree that the relationship information candidate is estimated to be the relationship information can also be used as a feature, and the accuracy of extracting the relationship information can be further improved.

  In addition, the said embodiment is only a specific example of this invention, and this invention is not limited only to the said embodiment. For example, according to the present invention, a program for realizing the functions shown in the configuration diagrams of FIGS. 1 and 9 or the procedures shown in the flows of FIGS. 2 and 10 are performed on a known computer via a recording medium or a communication line. It can also be realized by installing a program provided.

  Further, the related information candidate extraction unit 21 determines that the first morpheme of the phrase included in the input text corresponds to a predetermined functional verb (for example, “becomes”, “is”, “influence”, etc.) and immediately before the morpheme. When a morpheme corresponds to a particle (for example, “GA”, “O”, “NI”, etc.) associated with the functional verb, two clauses containing these morphemes are collected as one relation information candidate. It may be extracted. For example, the relationship information candidate extraction unit 21 stores a functional verb “effect” and a particle “wo” corresponding to the functional verb in a predetermined storage device (not shown). When the phrase “influence” and the phrase “influence” appear consecutively in the input text, the relationship information candidate extraction unit 21 summarizes the phrase “influence” and the phrase “influence”. “Influence” is extracted as a candidate for related information.

  In this case, since a phrase composed of a combination of a predetermined particle and a functional verb can be extracted as related information candidates, the number of related information candidates can be increased, and the extraction accuracy of optimum related information can be improved. it can.

  DESCRIPTION OF SYMBOLS 10 ... Analysis process part, 11 ... Morphological analysis part, 12 ... Dependency analysis part, 20 ... Relation information extraction process part, 21 ... Relation information candidate extraction part, 22 ... Central topic information extraction part, 25 ... Relation information extraction part, 26: Relationship estimation information acquisition unit, 40 ... Relationship estimation information storage unit

Claims (9)

An apparatus for extracting information related to a plurality of input proper expressions,
When the text including each of the specific expressions is input, an analysis processing unit that analyzes morphological analysis of the input text and a dependency of a clause constituting the input text;
When the analysis result by the analysis processing unit is obtained, after extracting a phrase including an independent word from other sentences different from the sentence including each specific expression from among a plurality of sentences constituting the input text, each specific expression And the noun phrases extracted in order from the beginning of the input text until all of the related information candidates are extracted, are stored in the storage area associated with the particle associated with the noun phrase, and all of the specific expressions and the related information candidates are stored. And a relation information extraction processing unit that acquires storage information indicating the storage state of noun phrases in the storage area and extracts relation information from relation information candidates based on the analysis result and the storage information. Related information extraction device characterized by the above. The relation information extraction processing unit acquires, as storage information, a particle in a storage area in which a noun phrase corresponding to the relation information candidate is stored when all of the specific expressions and relation information candidates are extracted. The related information extracting device according to claim 1. The relation information extraction processing unit acquires, as storage information, storage ranks in the storage area of noun phrases corresponding to the relation information candidates when all of the unique expressions and relation information candidates are extracted. The related information extracting device according to claim 1 or 2. The said relation information extraction process part will acquire the part of speech of the said relation information candidate as storage information, if all of each specific expression and the relation information candidate are extracted. Related information extraction device. When all of the specific expressions and the related information candidates are extracted, the related information extraction processing unit corresponds to any one of the specific expressions or the related information candidates among the noun phrases stored in the storage area. 5. The relation information extracting apparatus according to claim 1, wherein information indicating a storage state of a noun phrase having a dependency relationship with a phrase is acquired as storage information. When the analysis result obtained by the analysis processing unit is acquired, the related information extraction processing unit selects a phrase including an independent word from another sentence different from the sentence including each unique expression from among a plurality of sentences constituting the input text as a related information candidate And extracting the relationship estimation information indicating the degree to which the relationship information candidate is estimated as the relationship information and the storage information for each extracted relationship information candidate, and analyzing the result, the relationship estimation information and the storage. The relationship information extraction apparatus according to claim 1, wherein the relationship information is extracted from the relationship information candidates based on the information. A method of extracting information related to a plurality of input unique expressions using a computer,
When the text including each unique expression is input, the computer morphologically analyzes the input text and analyzes the dependency of the clause constituting the input text,
After extracting clauses containing independent words as related information candidates in other sentences that are different from the sentence containing each unique expression among the sentences that make up the input text, input until all of the specific expressions and related information candidates are extracted The noun phrases extracted in order from the beginning of the text are stored in the storage area associated with the particle associated with the noun phrase, and when all the specific expressions and candidate relational information are extracted, the noun phrases in the storage area are stored. A relation information extraction method characterized by acquiring storage information representing a storage state and extracting relation information from relation information candidates based on an analysis result and storage information.   The program for functioning a computer as each means of the related information extraction apparatus in any one of Claims 1 thru | or 6.   The program for making a computer perform each process of the relationship information extraction method of Claim 7.

Images