JP5488249B2 - Program and information processing apparatus - Google Patents

Description

  The present invention relates to a program and an information processing apparatus.

  In the field of natural language processing, techniques for identifying synonymous expressions that have the same meaning as a certain expression and paraphrased expressions that rephrase a certain expression are known.

  For example, in the technique described in Non-Patent Document 1, a bipartite graph connecting a certain word and its modifier or a modified word is created based on the dependency relationship of words in a sentence, and this bipartite graph is used. Thus, words having a dependency relationship similar to each other are specified as a paraphrase expression.

  Patent Document 1 discloses a technique for specifying a synonymous expression or paraphrased expression of a certain expression according to a preset rule. In Patent Document 1, when there are parentheses (“”, (), etc.) between a certain expression and another expression, and one expression is enclosed in parentheses, these expressions are referred to as paraphrase expressions. An example of a rule to do is described.

  Also, in Non-Patent Document 2, a rule for extracting relationships such as synonyms according to the syntax pattern of words in a sentence is defined, and words appearing in the syntax pattern defined by this rule are synonyms. A technique for specifying a word having a relationship such as is disclosed.

JP 2006-293731 A

Kazuhide Yamamoto, “Acquisition of lexical paraphrase knowledge from text”, The 8th Annual Conference of the Association for Natural Language Processing, pp. 639-642, March 2002 Marty A. Hearst, “Automatic Acquisition of Hyponyms from Large Text Corporation”, Proceedings of 14th International Conference of Computational Linguistics, 1992

  By the way, in expressions representing evaluations for some objects, phrases having different meanings may appear in sentences having the same syntax or similar syntax. In the expression representing evaluation, the same word may appear before and after words having different meanings. Therefore, if a candidate for an expression that is highly relevant to the expression representing the evaluation is identified with reference to the syntax of a sentence including the expression or words before and after the expression, expressions having different meanings are related to each other. May be identified as a candidate for a high expression.

  The present invention is more accurately related to expressions representing evaluations compared to a technique for identifying candidate expressions having high relevance by referring to the syntax of a sentence including a certain expression or words before and after the expression. It is an object of the present invention to provide a program and an information processing apparatus that specify high expression candidates.

  The invention according to claim 1 is an evaluation expression including a word representing an evaluation item for evaluating an evaluation object and a word representing an evaluation value for the evaluation item, and whether or not the evaluation expression is a positive expression. The evaluation expression storage means stored in association with each other is extracted, and the evaluation expression included in the processing target character string is extracted from the processing target character string, and the polarity of each of the extracted evaluation expressions is specified. A classification step for classifying the extracted evaluation expressions into one or more groups based on whether the evaluation object and the polarity are common between the extracted evaluation expressions, and the classification step. When there are a plurality of evaluation expressions classified into the same group, an output step of outputting the plurality of evaluation expressions to the related expression storage means in association with each other is executed on the computer A program for causing.

  The invention according to claim 2 is the invention according to claim 1, wherein, in the classification step, when the extracted evaluation expressions are arranged in the order of appearance in the processing target character string, the evaluation is performed between consecutive evaluation expressions. If the target and the polarity are common, the consecutive evaluation expressions are classified into the same group.

  The invention according to claim 3 is the invention according to claim 1 or 2, wherein in the output step, among the plurality of evaluation expressions, an evaluation expression in which words representing the evaluation items are in a common semantic class. No association between the evaluation expressions.

  The invention according to claim 4 is the invention according to any one of claims 1 to 3, wherein the evaluation expression is further added to the computer for each of the plurality of evaluation expressions to be output in the output step. An extraction step of extracting a character string including the character string to be processed from the character string to be processed, and the character string extracted in the extraction step is a word representing the evaluation item of the evaluation expression in the character string to be processed or the evaluation In the output step, character strings extracted for each of the plurality of evaluation expressions are associated with each other and output to the related expression storage unit.

  The invention according to claim 5 is the invention according to claim 4, wherein, in the output step, the character string including words in a common semantic class among the character strings extracted for each of the plurality of evaluation expressions, No association is made between the character strings.

  The invention according to claim 6 is the invention according to any one of claims 1 to 5, wherein the evaluation value is a numerical value among the evaluation items in the evaluation expression stored in the computer and further in the evaluation expression storage means. Reference value information storage means that associates and stores each word representing an evaluation item that can be expressed and the reference value of the evaluation value of the evaluation item, and the reference value information from the character string to be processed Based on the result of extracting the word representing the evaluation item stored in the storage means and the numerical value corresponding to the evaluation value of the evaluation item, and comparing the extracted reference value and the reference value associated with the extracted evaluation item , Causing the evaluation expression corresponding to the combination of the evaluation item and the numerical value and the polarity thereof to be specified in the evaluation expression storage means, and executing the second specifying step. Evaluation and further processed in the classification step expressions.

  The invention according to claim 7 is an evaluation expression including a word representing an evaluation item for evaluating the evaluation object and a word representing an evaluation value for the evaluation item, and whether or not the evaluation expression is a positive expression. The evaluation expression storage means stored in association with each other is extracted, and the evaluation expression included in the processing target character string is extracted from the processing target character string, and the polarity of each of the extracted evaluation expressions is specified. Classification means for classifying the extracted evaluation expressions into one or more groups based on whether the evaluation object and the polarity are common between the evaluation expressions extracted by the specifying means; and And an output means for associating the plurality of evaluation expressions with each other and outputting them to the related expression storage means when there are a plurality of evaluation expressions classified into the same group by the classification means. It is a processing apparatus.

  According to the invention according to claim 1 or 7, the evaluation is performed more accurately compared to a technique for identifying a highly relevant expression candidate with reference to a sentence syntax including an expression or words before and after the expression. It is possible to specify expressions that are highly relevant to the expression that represents.

  According to the second aspect of the present invention, evaluation expressions representing positive evaluation or negative evaluation for the same evaluation object can be output in association with each other successively in the order of appearance in the character string to be processed.

  According to the third aspect of the present invention, it is possible not to associate evaluation expressions in which words representing evaluation items are in a common semantic class.

  According to the invention of claim 4, a character including each of a plurality of evaluation expressions associated with each other and a phrase having a dependency relationship between a word representing an evaluation item of the evaluation expression or a word representing an evaluation value A column can be extracted from a character string to be processed, and the extracted character strings can be associated with each other and output.

  According to the invention which concerns on Claim 5, the character string containing the word which exists in a mutually common semantic class can be made not to associate.

  According to the invention which concerns on Claim 6, the expression candidate with high relevance can be specified about the evaluation expression containing the evaluation item by which evaluation is represented numerically.

It is a block diagram which shows the example of the outline of an internal structure of information processing apparatus. It is a figure which shows the example of the data content of an evaluation expression dictionary. It is a figure which shows the example of the text of a process target. It is a figure for demonstrating the example of grouping of the evaluation expression extracted from the text of a process target. It is a figure which shows the example of the data content of a related expression memory | storage part. It is a flowchart which shows the example of the procedure of the process which information processing apparatus performs. It is a figure for demonstrating the other example of grouping of the evaluation expression extracted from the text of a process target. It is a figure which shows the other example of the text of a process target. It is a figure which shows the example of the information of a reference value. It is a block diagram which shows the example of the hardware constitutions of a computer.

  In the example of the embodiment of the present invention, by analyzing the text (sentence) to be processed, a paraphrase expression candidate for an expression that evaluates some evaluation object is specified. Here, the “paraphrased expression” of a certain expression refers to an expression that can be replaced with the certain expression because it has some relation to the certain expression. For example, the “paraphrase expression” of a certain expression may be an expression having the same meaning as the certain expression, or may be an expression having a similar meaning to the certain expression. Further, in the example of the present embodiment, one expression may be specified as a candidate for the other paraphrase expression, even if the expressions are not necessarily recognized as being replaceable depending on the knowledge and way of thinking of each person. For example, if a person who does not have knowledge in the field of specialization can help to understand the evaluation by rephrasing the expression of the evaluation in one field of expertise, it is not necessarily a replacement for the expert in the field of expertise. Even expressions that are not considered possible may be candidates for paraphrased expressions. Therefore, in the example of the embodiment of the present invention, a candidate for a paraphrased expression of a certain word is regarded as an expression highly relevant to the certain word. Hereinafter, an expression that is a candidate for a paraphrase expression of a certain expression, that is, an expression that is highly relevant to a certain expression is referred to as a “related expression”.

  FIG. 1 is a block diagram showing an outline of an internal configuration of an information processing apparatus according to an example of an embodiment of the present invention. The information processing apparatus 10 includes a reference data storage unit 110, a corpus analysis unit 120, an evaluation expression extraction unit 130, an evaluation expression classification unit 140, a related expression generation unit 150, a semantic class determination unit 160, an output processing unit 170, and a related expression storage Part 180.

  The reference data storage unit 110 stores various data used in analyzing the text to be processed. The reference data storage unit 110 includes an analysis dictionary 112, an evaluation expression dictionary 114, and a semantic dictionary 116.

  The analysis dictionary 112 stores a word and information indicating a grammatical role of the word in association with each other and stores a grammar rule of a language (Japanese in this example) in which a text to be processed is described. It is. In the analysis dictionary 112, not only general words but also technical terms used in various specialized fields may be registered. For example, when a sentence in a specific specialized field is to be processed, technical terms in the specific specialized field are registered in the analysis dictionary 112 in advance. In the example of the present embodiment, it is assumed that a sentence in the medical field is a processing target of the information processing apparatus 10 and technical terms in the medical field are registered in the analysis dictionary 112 together with general words.

  The evaluation expression dictionary 114 associates and stores an evaluation expression representing an evaluation with respect to some evaluation object and a polarity indicating whether the evaluation expression is a positive evaluation or a negative evaluation. The evaluation expression dictionary 114 of the example of this embodiment stores evaluation expressions in the medical field and their polarities. FIG. 2 shows an example of data contents of the evaluation expression dictionary 114.

  FIG. 2 shows an example of an evaluation expression for evaluating a change in the state of the evaluation target. In the example of FIG. 2, in the evaluation expression dictionary 114, the polarity of the evaluation expression is registered in association with each evaluation expression including a set of “object”, “attribute”, and “evaluation value” of evaluation. The “object” of the evaluation expression represents the evaluation object in the evaluation expression. In the evaluation expressions illustrated in FIG. 2, “liver” is an evaluation target. The “attribute” of the evaluation expression is a word representing an evaluation item for evaluating the evaluation object. FIG. 2 shows an example of an evaluation expression including the words “liver function”, “GOP”, and “GPT” as attributes. The “evaluation value” of the evaluation expression is a word that represents the evaluation of the corresponding attribute. FIG. 2 shows an example of an evaluation expression including the words “improvement”, “deterioration”, “decrease”, “decrease”, “increase”, and “increase” as evaluation values. Each of the evaluation values illustrated in FIG. 2 is a word that represents a change in the state of the evaluation target. The “polarity” associated with each evaluation expression represents whether the evaluation expression is positive or negative. In the table of FIG. 2, the polarity value “positive” indicates that the evaluation expression is positive, and the polarity value “negative” indicates that the evaluation expression is negative.

  The data contents of the evaluation expression dictionary 114 are generated and registered in advance before each element described later of the information processing apparatus 10 starts processing. For example, the evaluation expression including the evaluation object, the attribute, and the evaluation value is extracted from a plurality of sentences in the field related to the text to be processed (the medical field in the example of the present embodiment), and the polarity is determined. By registering the expression and the polarity in association with each other in the evaluation expression dictionary 114, the data content of the evaluation expression dictionary 114 is generated. The extraction of the evaluation expression and the determination of the polarity may be performed using a conventionally known technique. For example, JP 2005-235014 A and Reference 1 (Nozomi Kobayashi, Kentaro Inui, Yuji Matsumoto, Kenji Tateishi, Shunichi Fukushima, “Collecting Evaluation Expressions for Opinion Extraction”, Natural Language Processing, 12 (2) , 2005) may be used.

  Referring again to FIG. 1, the semantic dictionary 116 stores information regarding the meaning of common words and technical terms. In the example of this embodiment, the semantic dictionary 116 includes a thesaurus for common words and technical terms. The thesaurus is a systematic dictionary that classifies words according to upper / lower relations, partial / whole relations, synonymous relations, synonymous relations, etc., and has a hierarchical structure according to the classification of the words. The semantic dictionary 116 in the example of the present embodiment further includes a synonym dictionary that represents synonyms for each word.

  Returning to the description of FIG. 1, the corpus analysis unit 120 refers to the analysis dictionary 112 and performs morphological analysis and syntax analysis on the text to be processed. Morphological analysis and syntax analysis may be performed using a method conventionally known in the natural language processing technology. The analysis by the corpus analysis unit 120 is preprocessing for processing in each unit described later.

  Note that the text to be processed is acquired from, for example, a database (not shown) that stores documents. In the example of this embodiment, the text to be processed is acquired from a document in a database storing documents (such as academic papers and reports on patient medical conditions) created by doctors, nurses, and staff of medical institutions. And All sentences included in all documents in the database may be text to be processed, or all sentences included in documents specified by the user among documents in the database may be text to be processed. Alternatively, a sentence included in one or more portions designated by the user in one or more documents in the database may be set as the text to be processed.

  The evaluation expression extraction unit 130 refers to the evaluation expression dictionary 114 and extracts an evaluation expression from the text to be processed analyzed by the corpus analysis unit 120. In the following, the text “GPT-dominated deterioration of liver function was observed. GOP increased, GPT increased significantly. Immediate improvement of liver function was confirmed by fluid replacement etc.” is shown in FIG. Taking the case of the text to be processed as an example, how the evaluation expression extraction unit 130 extracts the evaluation expression will be described. In this example, it is assumed that the data contents illustrated in FIG. 2 are registered in the evaluation expression dictionary 114.

  The evaluation expression extraction unit 130 specifies a word representing an attribute and a word representing an evaluation value included in the evaluation expression registered in the evaluation expression dictionary 114 in the text to be processed. In FIG. 3, words surrounded by broken-line squares are examples of attributes and evaluation value words specified by the evaluation expression extraction unit 130. Furthermore, the evaluation expression extraction unit 130 specifies an attribute that is paired with the specified evaluation value. The attribute corresponding to the specified evaluation value may be specified based on, for example, the dependency relationship of the text to be processed, or Reference 2 (Ryu Iida, Nozomi Kobayashi, Kentaro Inui, Yuji Matsumoto, Kenji Tateishi , Shunichi Fukushima, “Attribute-Evaluation Pair Identification by Machine Learning for Opinion Extraction”, Natural Language Processing Study Group of Information Processing Society of Japan, 2005-NL-165). May be specified. In the technique described in Reference Document 2, the problem of extracting an evaluation expression (called “opinion” in the document) represented by a triple of <object, attribute, evaluation value> from the text to be processed, (1) The problem of identifying the <attribute, evaluation value> pair, and (2) Whether the identified pair has opinionality (whether the condition that expresses what description is regarded as an opinion) The evaluation expression is extracted by analyzing using a method based on machine learning. In FIG. 3, the attribute and the evaluation value that form a pair are connected by a broken-line arrow. The evaluation expression extraction unit 130 extracts a set of a word representing the attribute specified in the text to be processed and a word representing the evaluation value. In the example of FIG. 3, (liver function, deterioration), (GOP, increase), (GPT, increase), and (liver function, improvement) are extracted as a set of (attribute, evaluation value).

  In this example with reference to FIG. 3, the evaluation expression extraction unit 130 acquires the evaluation target in the evaluation expression separately from the text to be processed. For example, the evaluation expression extraction unit 130 acquires a document including the text to be processed from the database, analyzes the document, specifies a character string representing the subject of the document, and stores the document in the evaluation expression dictionary 114 from this character string. Extract words representing the evaluation target. The character string representing the subject of the document may be, for example, the name of the document. Further, for example, when the document is a report on a patient's medical condition, a character string describing the patient's disease name may be a character string representing the subject of the document. Also, instead of specifying the character string representing the subject of the document, the character string corresponding to the title of the part (chapter, section, item, etc.) containing the text to be processed in the document is specified and evaluated from this character string. A word representing the target may be extracted. In the above example, the character string that represents the evaluation target is extracted from what character string in the document, and the character string that is the basis of the extraction of the word that represents the evaluation target can be present in the document. The information is set in advance and stored in the reference data storage unit 110. With reference to this information, the evaluation expression extraction unit 130 may perform document analysis and evaluation target extraction. In another example, designation of text to be processed by the user may be accepted and designation of an evaluation target may be accepted from the user. The evaluation expression extraction unit 130 associates the evaluation object acquired as in each of the above-described examples with a set of an attribute and an evaluation value extracted from the text to be processed, and includes (object, attribute, evaluation value). The evaluation expression consists of a set. In the example of FIG. 3, an evaluation target “liver” is acquired, and evaluation expressions (liver, liver function, deterioration), (liver, GOP, increase), (liver, GPT, increase), (liver, liver function, improvement) Is obtained.

  Further, the evaluation expression extraction unit 130 acquires, for each of the extracted evaluation expressions, a polarity value registered in the evaluation expression dictionary 114 in association with the evaluation expression. FIG. 4 illustrates an evaluation expression (object, attribute, evaluation value) extracted from the text to be processed in the example of FIG. 3 and the polarity value acquired for each evaluation expression. The polarity value of each evaluation expression shown in the table of FIG. 4 is a polarity value associated with the evaluation expression in the evaluation expression dictionary 114 illustrated in FIG.

  Returning to the description of FIG. 1, the evaluation expression classifying unit 140 divides the extracted evaluation expressions into one or more groups based on whether or not the extracted evaluation expressions have the same attribute word and polarity. Classify. A plurality of evaluation expressions classified into the same group are candidates for related expressions.

  A specific example of classification by the evaluation expression classification unit 140 will be described with reference to FIG. The table of FIG. 4 is a table in which the evaluation expressions extracted from the text to be processed in the example of FIG. 3 and their polarities are arranged from the upper line to the lower line in the order of appearance of the evaluation expressions in the text to be processed. It is. In this example, the evaluation expression classifying unit 140 classifies the evaluation expressions that appear in succession in the text to be processed and that have the same evaluation object and polarity into the same group. When evaluation expressions of the same polarity are described consecutively for the same evaluation object, it is recognized that the attitude of the person who made the description to the evaluation object is consistent. Therefore, there is a possibility that such evaluation expressions express the same evaluation in other words for the same evaluation object. Therefore, the evaluation expression classifying unit 140 classifies such evaluation expressions into the same group as related expression candidates. Referring to FIG. 4, the first three evaluation expressions that are “liver” as the evaluation target and consecutive in the negative as “negative” are classified into group 1, and the evaluation target is “liver” that is the same as group 1 However, the fourth evaluation expression having a polarity of “positive” and different is classified into group 2.

  Referring again to FIG. 1, the related expression generation unit 150 generates a set of related expressions using the result of classification by the evaluation expression classification unit 140. For example, for the group including a plurality of evaluation expressions among the groups in which the evaluation expression classification unit 140 classifies the evaluation expressions, the related expression generation unit 150 processes a character string including each of the evaluation expressions included in the group. Extract from column. Here, the character string extracted for each evaluation expression includes a phrase having a dependency relationship with a word representing an attribute of the evaluation expression or a word representing an evaluation value in the text to be processed. For example, in the case of the evaluation expression (liver, liver function, deterioration) included in the group 1 in the example of FIG. 4, from the text to be processed (FIG. 3), the attribute “liver function” and the evaluation value “deterioration” are included. Character strings such as “deterioration of liver function” and “deterioration of liver function dominant in GPT” are extracted. In the case of the evaluation expression (liver, GOP, increase), the character string “Increase in GOP” is extracted. In the case of the evaluation expression (liver, GPT, increase), the character strings “Increase in GPT” and “Significant GPT” are extracted. "Near rise" is extracted. It should be noted that the related expression generation unit 150 has a minimum unit character string (character string obtained by the shortest path in the syntax) syntactically established as a phrase or sentence among the character strings including the evaluation expression attribute and the evaluation value. May be extracted from the text to be processed. In this case, the character strings “deterioration of liver function”, “increase in GOP”, and “increase in GPT” are extracted from the text to be processed.

  The related expression generation unit 150 sets two of the character strings extracted for each evaluation expression included in a certain group as a set of related expressions. However, character strings that are determined to belong to a common semantic class between attributes representing attributes in the evaluation expression are not included in the set of related expressions. Further, even in a character string extracted from the processing target character string with respect to the evaluation expression, character strings including words determined to belong to a common semantic class are not used as a set of related expressions. Here, the semantic class is a set of words grouped by a common semantic element in a thesaurus dictionary or the like, and the semantic element is a basic unit of meaning used when the meaning is handled by a computer. For example, if the semantic element is “animal”, the semantic class is a set of words such as “dog”, “horse”, and “monkey”. The related expression generation unit 150 requests the semantic class determination unit 160 to determine whether there are words in the same semantic class among the words representing the attributes of the evaluation expression included in a certain group, and the evaluation expression It is determined whether or not there is a character string that includes words in a common semantic class among the character strings extracted for.

  The semantic class determination unit 160 receives words from the related expression generation unit 150 and determines words in a common semantic class. When a plurality of words representing different matters appear in an evaluation expression for evaluating a certain evaluation object, it can be considered that these words have a common semantic class. For example, in the example described above with reference to FIGS. 3 and 4, the attributes “GOP” and “GPT” of the evaluation expression represent different inspection items. Therefore, in this example, “GOP” and “GPT” are determined to be words in a common semantic class. The semantic class determination unit 160 in the example of this embodiment refers to the semantic dictionary 116 to determine whether words are in a common semantic class. More specifically, the semantic classes in the thesaurus included in the semantic dictionary 116 are the same, and words that are not registered as synonyms in the synonym dictionary included in the semantic dictionary 116 are defined as common semantic classes. Judge that there is. Here, whether or not the semantic classes of a plurality of words are the same may be determined according to the hierarchical structure of the concept of the thesaurus. For example, in the hierarchical structure of the thesaurus, when the concepts positioned immediately above the concepts (items) corresponding to a plurality of words are the same, the semantic classes may be determined to be the same. In other words, when nodes corresponding to a plurality of words have the same parent node in a tree structure corresponding to the thesaurus hierarchical structure, the semantic classes may be determined to be the same. In the case of the above-mentioned examples of “GOP” and “GPT”, these words belong to the immediate subordinate concept of the same “inspection value” concept (semantic class) and are not synonymous in the thesaurus hierarchy. , It is determined to be in a common semantic class.

  The output processing unit 170 performs processing for outputting a set of related expressions generated by the related expression generating unit 150. For example, the generated set of related expressions is output to the related expression storage unit 180. By this output processing, a set of related expressions is registered in the related expression storage unit 180. Further, the output processing unit 170 may perform a process of displaying a set of related expressions on a display device (not shown).

  The related expression storage unit 180 stores a set of related expressions generated by the related expression generation unit 150 and output by the output processing unit 170. FIG. 5 shows an example of the data contents of the related expression storage unit 180. FIG. 5 shows an example of a set of related expressions generated by the related expression generation unit 150 when the evaluation expression of FIG. 4 is extracted from the processing target text of FIG. Two expressions described in one line of the table of FIG. 5 represent a set of related expressions. Referring to FIG. 5, the expression “deterioration of liver function” based on the evaluation expression (liver, liver function, deterioration) and the expression “increase GOP” based on the evaluation expression (liver, GOP, increase) are related expressions. It is registered as a group. Also, “deterioration of liver function” and expressions “elevation of GPT” and “significant increase of GPT” based on evaluation expressions (liver, GPT, increase) are registered in association with each other. Furthermore, the expressions “deterioration of liver function predominantly GPT” based on the evaluation expression (liver, liver function, deterioration) and “increased GPT” and “GPT” are registered in association with each other. It should be noted that expressions based on evaluation expressions (liver, GOP, increase) and (liver, GPT, increase) determined by the semantic class determination unit 160 that the word representing the attribute is in a common semantic class are a set of related expressions. Not registered as. In addition, the expression “deterioration of liver function with GPT predominance” includes the word “GPT”, and thus is not associated with “increased GOP” including the word “GOP” in a common semantic class.

  Hereinafter, an example of a procedure of processing performed by the information processing apparatus 10 will be described with reference to FIG. For example, when the user instructs to start processing using an input device (not shown), the information processing apparatus 10 starts processing of the procedure in the example of FIG.

  First, the information processing apparatus 10 acquires a text to be processed (step S10). In this example, the text to be processed is acquired from the document in the database storing the medical field document.

  The corpus analysis unit 120 of the information processing apparatus 10 performs morphological analysis and syntax analysis on the text to be processed (step S12). In step S12, the analysis dictionary 112 is referred to.

  After step S12, the evaluation expression extraction unit 130 extracts an evaluation expression from the text to be processed (step S14). As in the example described above with reference to FIG. 3, the evaluation expression extraction unit 130 refers to the evaluation expression dictionary 114 and extracts a combination of a word representing an attribute and a word representing an evaluation value from the text to be processed. . Furthermore, the evaluation target in the evaluation expression in the processing target text is acquired from the document containing the processing target text or as specified by the user, and the acquired evaluation target corresponds to the combination of the extracted attribute and evaluation value. Then, an evaluation expression represented by (evaluation object, attribute, evaluation value) is obtained.

  The evaluation expression extraction unit 130 specifies each polarity of the evaluation expression extracted in step S14 (step S16). The evaluation expression extraction unit 130 of this example acquires the value of the polarity registered in the evaluation expression dictionary 114 in association with the evaluation expression for each of the extracted evaluation expressions. The evaluation expression extraction unit 130 passes the results of steps S14 and S16 to the evaluation expression classification unit 140. FIG. 4 described above is an example of evaluation expressions and their polarities obtained as a result of steps S14 and S16.

  The evaluation expression classification unit 140 classifies the evaluation expressions extracted in step S14 into one or more groups based on whether or not the evaluation object and polarity are common between these evaluation expressions (step S18). As in the example described above with reference to FIG. 4, the evaluation expression classifying unit 140 continues these evaluation expressions when the evaluation object and the polarity are common among the evaluation expressions that are consecutive in the order of appearance in the text to be processed. Classify evaluation expressions into the same group. In the example of FIG. 4, the evaluation target is “liver” in all of the extracted evaluation expressions, and therefore, if the same polarity is continuous, they are classified into the same group.

  As another example of the classification result in step S18, FIG. 7 shows an example in the case where there is an extracted evaluation expression including different evaluation objects. Referring to the table of FIG. 7, it is assumed that three evaluation expressions for the object A, four evaluation expressions for the object B, and one evaluation expression for the object C are extracted in step S14. Also, it is assumed that these evaluation expressions appear in the text to be processed in order from the top row in the table of FIG. At this time, two evaluation expressions in which the polarity “positive” continues in the object A are classified into the group a, and the evaluation expression (polarity “negative”) of the next object A is classified into the group b alone. Further, two evaluation expressions having the polarity “negative” continuous in the object B are classified into the group c, and two evaluation expressions having the polarity “positive” in the object B are classified into the group d, and the evaluation expressions of the remaining object C are Furthermore, it is classified into another group e.

  The evaluation expression classification unit 140 passes the result of the classification in step S18 to the related expression generation unit 150.

  Next, the related expression generation unit 150 causes the semantic class determination unit 160 to determine whether there is a word in a common semantic class among the words representing the attributes of the evaluation expression for each group (step S20). For example, for the group including a plurality of evaluation expressions, the related expression generation unit 150 determines whether there is a common class among words representing the attributes of the evaluation expressions included in the group. The determination unit 160 makes a determination. In the case of the group 1 example in FIG. 4, “liver function” and “GOP”, “liver function” and “GPT”, and “GOP” and “GPT” are in a common semantic class. Whether or not the semantic class determination unit 160 determines. In this example, as described above, the semantic class determination unit 160 determines that “GOP” and “GPT” are in a common semantic class, and “liver function”, “GOP”, and “liver function”. And “GPT” are determined not to be in a common semantic class.

  After determining whether or not they are in a common semantic class, the related expression generation unit 150 extracts a character string including each evaluation expression of each group including a plurality of evaluation expressions from the text to be processed (step S22). In the case of group 1 in FIG. 4, in step S22, as described above, “deterioration of liver function”, “deterioration of liver function superior to GPT”, “increase in GOP”, “GPT” from the processing target text in FIG. "Increase" and "Significant increase in GPT" are extracted. The related expression generation unit 150 sets two of the character strings extracted for each evaluation expression of the same group as a set of related expressions. However, character strings including words determined to be in the common semantic class in step S20 are not set as related expressions. The related expression generation unit 150 passes the generated set of related expressions to the output processing unit 170.

  After step S22, the output processing unit 170 registers the set of related expressions received from the related expression generation unit 150 in the related expression storage unit 180 (step S24). FIG. 5 described above is an example of the registration result in step S24. After step S24, the process of the procedure in the example of FIG. 6 ends.

  By the processing of the procedure in the example of FIG. 6, a set of related expressions, that is, a set of expression candidates that can be paraphrased (highly related to each other) is registered in the related expression storage unit 180. The related expression storage unit 180 is used for sentence search processing and the like. For example, the related expression of the expression included in the search query input by the user is acquired from the related expression storage unit 180, and not only the expression included in the input search query but also the related expression of this expression is used as a search key as a document. It is conceivable to perform processing such as searching.

  As described above, in the example described with reference to FIGS. 3 to 5, both the attribute of the evaluation expression and the evaluation value are included in one sentence in the text to be processed. In another example of the text to be processed, the evaluation expression attribute and the evaluation value may appear in different sentences. For example, referring to FIG. 8, the text to be processed is “Platelet survey results were obtained. A marked decrease was observed. An increase in HHV-6 viral load was observed. Reactivation of HHV-6 was observed. It is possible that he was involved. ” A broken-line square and a broken-line arrow in FIG. 8 indicate a set of a word representing an attribute of the evaluation expression and a word representing the evaluation value, as in FIG. In FIG. 8, (platelet, decrease), (HHV-6 viral load, increase), and (HHV-6, reactivation) are extracted as a set of evaluation expression attributes and evaluation values, and these three evaluations are extracted. Assume that the expressions are classified into the same group in step S18 of FIG. In the set of (platelet, decrease), the word “platelet” representing an attribute and the word “decrease” representing an evaluation value appear in two different sentences. In such a case, as a character string including (platelet, decrease), for example, “Platelet survey results were obtained. Decrease was observed.” And “Platelet survey results were obtained. Is recognized, and the strings “HHV-6 viral load”, “HHV-6 viral load, increase”, “HHV-6 viral load”, “HHV-6 viral load”, “reactivation of HHV-6” Or the like as a related expression.

  The example of embodiment described above is only an example of embodiment of this invention, and there may be various modifications.

  In one modification, for an attribute whose evaluation value can be represented by a numerical value among the attributes of the evaluation expression, when a numerical value representing the evaluation value of the attribute is described in the text to be processed, the evaluation expression extracting unit 130 A pair of a word and a numerical value representing the attribute may be extracted from the text to be processed. In this modified example, the evaluation expression dictionary 114 or the semantic dictionary 116 further stores information indicating the reference value of the evaluation value of the attribute (evaluation item) in which the evaluation value can be expressed by a numerical value in the evaluation expression. For example, various examination items in medicine may be attributes of evaluation expressions, and evaluation values are represented by numerical values. Therefore, the evaluation expression dictionary 114 or the semantic dictionary 116 of the present modification stores the word representing the inspection item among the attributes in the evaluation expression in association with the information representing the reference value of the inspection value of the inspection item. FIG. 9 shows an example of such reference value information. In the table in the example of FIG. 9, a word representing an inspection item whose evaluation value can be expressed as a numerical value among attributes of the evaluation expression and a reference value of the evaluation value are associated with each other. The reference value information as illustrated in FIG. 9 is generated in advance using, for example, examination items of a test that can be performed in a medical institution and the reference value, and is registered in the evaluation expression dictionary 114 or the semantic dictionary 116.

  When the reference value information of the example of FIG. 9 is registered in the evaluation expression dictionary 114 in addition to the data contents of the example of FIG. 2, for example, the sentence “γ-GTP becomes 110, and a decrease in liver function is observed. ”Is the text to be processed. At this time, the evaluation expression extraction unit 130 extracts (γ-GTP, 110) and (liver function, decline) as a set of attributes and evaluation values. Note that “liver” is acquired as an evaluation target. The evaluation expression (liver, liver function, decline) extracted from the processing target text is the evaluation expression itself registered in the evaluation expression dictionary 114, and its polarity is “negative”. The evaluation expression extracting unit 130 refers to the reference value information in the example of FIG. 9 for the extracted attribute and numerical value pair (γ-GTP, 110), and the reference value “50 or less” of “γ-GTP”. Is compared with the extracted numerical value “110”. As a result of this comparison, since the extracted numerical value “110” is larger than the reference value, for example, among the evaluation expressions including the attribute “γ-GTP”, an evaluation value “increase” representing a change in which the numerical value increases or An evaluation expression (liver, γ-GTP, increase) or (liver, γ-GTP, increase) including “increased” is selected as an evaluation expression corresponding to (γ-GTP, 110). Then, the polarity of the selected evaluation expression is acquired from the evaluation expression dictionary 114, and this evaluation expression and its polarity are evaluated together with other evaluation expressions (liver, liver function, decline) extracted from the text to be processed and its polarity. It is a target of classification processing by the expression classification unit 140. When the related expression generation unit 150 performs a process of extracting a character string from the text to be processed for the evaluation expression corresponding to the combination of the attribute and the numerical value, the related expression generating unit 150 includes a character string (for example, “γ -GTP may be 110 "), or a character string in which the numerical value is replaced with the word of the evaluation value of the corresponding evaluation expression (eg," γ-GTP increases ") may be acquired.

  In this modification example with reference to FIG. 9, the evaluation expression corresponding to the combination of the attribute and the numerical value extracted from the text to be processed is a comparison between the extracted numerical value and the reference value associated with the attribute. Based on the result of For example, if the extracted numerical value is larger than the reference value, it corresponds to an evaluation expression that includes a word (increase, increase, etc.) that represents a change that increases the numerical value as an evaluation value and includes the attribute extracted as a pair with the numerical value. The evaluation expression to be used. On the contrary, if the extracted numerical value is smaller than the reference value, an evaluation expression including a word (decrease, decrease, etc.) indicating a change in the numerical value as an evaluation value and an attribute extracted as a pair with the numerical value. The corresponding evaluation expression. Further, if the extracted numerical value is the same as the reference value, there may be no corresponding evaluation expression. In addition, when the reference value of an attribute is represented by a numerical value range, if the extracted numerical value exceeds the upper limit of the reference value range, the extracted numerical value includes an evaluation expression that includes a word representing a change in the numerical value. If the value is below the lower limit of the range of the reference value, an evaluation expression representing a change in which the numerical value becomes smaller may be set as a corresponding evaluation expression.

  In the embodiment and the modification described above, evaluation expressions (including evaluation values representing changes such as increase, increase, decrease, decrease, etc.) for evaluating changes in the state of the evaluation target are registered in the evaluation expression dictionary 114. In another modification, not only an evaluation expression for evaluating a change in state but also an evaluation expression for evaluating the state of the evaluation object itself may be registered in the evaluation expression dictionary. For example, an evaluation expression that includes a word that simply describes the state as an evaluation value instead of a word that represents a change over time, such as “good”, “bad”, “high”, “low”, “large”, “small”, etc. Further, it may be registered in the evaluation expression dictionary 114. In this case, in the above-described modified example in which the evaluation expression including the attribute whose evaluation value is represented by a numerical value is processed, the magnitude relationship between the numerical value extracted from the text to be processed and the reference value associated with the corresponding attribute is The evaluation expression including the corresponding word as the evaluation value may be an evaluation expression corresponding to the extracted numerical value and attribute pair. For example, when the extracted numerical value is larger than the reference value, an evaluation expression including the word “high” is used. When the extracted numerical value is smaller than the reference value, the evaluation expression including the word “low” is An evaluation expression corresponding to a pair with an attribute. If the extracted numerical value is the same as the reference value, it is determined that there is no evaluation expression corresponding to the combination of the numerical value and the attribute.

  Further, in the above description, the evaluation expression classification unit 140 classifies the evaluation objects that have the same evaluation object and polarity among the evaluation expressions that are consecutive in the order of appearance in the text to be processed into the same group. In one modification, the evaluation expression classification unit 140 may simply classify evaluation expressions having the same evaluation object and polarity into the same group without considering the appearance order in the text to be processed. In this case, even if the evaluation expressions are not consecutive in the order of appearance in the text to be processed, they are classified into the same group as long as the evaluation object and polarity are common. In still another example, classification may be performed by further using conjunctions and modalities in the text to be processed, or classification as a clustering problem based on machine learning.

  Moreover, in the above description, the semantic class determination unit 160 determines that words having the same semantic class in the thesaurus and not synonyms are different words in a common semantic class. In one variation, a phrase including a word representing an attribute in a text to be processed has a juxtaposed structure using a conventionally known parsing technique (for example, “an increase in GOP, a significant increase in GPT). ", Two phrases are juxtaposed), and the detection result and determination of semantic classes in the thesaurus may be combined to determine the presence or absence of different words in a common semantic class.

  In the above description, the related expression generation unit 150 extracts a character string including the evaluation expression from the text to be processed, and associates the extracted character strings with each other and registers them in the related expression storage unit 180. In one modification, the extracted evaluation expression itself may be further registered in the related expression storage unit 180. For example, referring to FIG. 4, words representing attributes for three evaluation expressions (liver, liver function, deterioration), (liver, GOP, increase), and (liver, GPT, increase) classified into the same group 1 are shown. In the related expression storage unit 180, (liver, liver function, deterioration), (liver, GOP, increase), (liver, liver function, deterioration) and (liver, GPT, increase), which are not in a common semantic class, are stored. You may register. Further, when registering the evaluation expression itself in the related expression storage unit 180, a process of extracting a character string including the evaluation expression from the text to be processed and associating the extracted character string with the associated expression and registering it in the related expression storage unit 180 is omitted. May be.

  Note that the related expression generation unit 150 does not necessarily have to explicitly generate a “set” of related expressions. A plurality of evaluation expressions that are classified into the same group by the evaluation expression classification unit 140 and do not include words in a common semantic class (and character strings extracted from the text to be processed based on each evaluation expression) May be registered in the related expression storage unit 180 in association with each other.

  In the above, an evaluation expression composed of a set of three types of values, that is, an object, an attribute, and an evaluation value is registered in the evaluation expression dictionary 114. In one modified example, an evaluation expression including a word representing an attribute and a word representing an evaluation value is registered in the evaluation expression dictionary 114 without including the evaluation target, and the attribute and the evaluation extracted from the text to be processed are evaluated. You may perform similarly the process of the above-mentioned embodiment and various modifications by making a set with a value into evaluation expression. In the case of this example, an evaluation target may be specified from a word representing an attribute. For example, information that associates each evaluation object with a word that represents the attribute is registered in the evaluation expression dictionary 114 or the semantic dictionary 116, and the evaluation object associated with the word that represents the attribute extracted from the text to be processed is stored. , And specified as an evaluation target of an evaluation expression including the attribute.

  In the above description, an example in which medical text is a text to be processed has been described. Of course, the processing of the example of the present embodiment may be performed in the same manner as described above when a sentence in another specialized field is used as a text to be processed. Alternatively, a general sentence that is not limited to a specialized field may be used as a text to be processed.

  The information processing apparatus 10 exemplified above is typically realized by executing a program describing functions or processing contents of each unit of the information processing apparatus 10 described above on a general-purpose computer. In the computer, for example, as shown in FIG. 10, a CPU (central processing unit) 80, a memory (primary storage) 82, various I / O (input / output) interfaces 84, and the like are connected via a bus 86 as hardware. Circuit configuration. In addition, a hard disk drive (HDD) 88, a disk drive 90 for reading various types of portable non-volatile recording media such as a CD, a DVD, and a flash memory is connected to the bus 86 via, for example, an I / O interface 84. Connected. Such a drive 88 or 90 functions as an external storage device for the memory. A program in which the processing content of the embodiment is described is stored in a fixed storage device such as the HDD 88 via a recording medium such as a CD or DVD or via a network, and is installed in a computer. The program stored in the fixed storage device is read into the memory and executed by the CPU, whereby the processing of the embodiment is realized.

  In the above, the embodiment of the example in which the information processing apparatus 10 is realized by one computer has been described. However, the functions of various examples of the information processing apparatus 10 described above may be realized by being distributed to a plurality of computers. Good.

  DESCRIPTION OF SYMBOLS 10 Information processing apparatus, 80 CPU, 82 Memory, 84 I / O interface, 86 bus, 88 HDD, 90 Disk drive, 110 Reference data storage part, 112 Analysis dictionary, 114 Evaluation expression dictionary, 116 Semantic dictionary, 120 Corpus analysis part , 130 evaluation expression extraction unit, 140 evaluation expression classification unit, 150 related expression generation unit, 160 semantic class determination unit, 170 output processing unit, 180 related expression storage unit.

Claims (7)

Associating an evaluation expression including a word representing an evaluation item for evaluating an evaluation object and a word representing an evaluation value for the evaluation item, and a polarity indicating whether or not the evaluation expression is a positive expression A step of referring to the stored evaluation expression storage means, extracting the evaluation expression included in the character string to be processed from the character string to be processed, and specifying the polarity of each of the extracted evaluation expressions;
A classification step of classifying the extracted evaluation expressions into one or more groups based on whether the evaluation object and the polarity are common among the extracted evaluation expressions;
When there are a plurality of evaluation expressions classified in the same group in the classification step, an output step of outputting the plurality of evaluation expressions to the related expression storage means in association with each other;
A program that causes a computer to execute. In the classification step, when the extracted evaluation expressions are arranged in the order of appearance in the character string to be processed, if the evaluation object and the polarity are common among consecutive evaluation expressions, the continuous evaluation expressions Classify them into the same group,
The program according to claim 1. In the output step, among the plurality of evaluation expressions, for the evaluation expressions in which the words representing the evaluation items are in a common semantic class, the association between the evaluation expressions is not performed.
The program according to claim 1 or 2, characterized in that The computer is further caused to execute an extraction step of extracting a character string including the evaluation expression from the character string to be processed for each of the plurality of evaluation expressions to be output in the output step,
The character string extracted in the extraction step includes a word having a dependency relationship with a word representing the evaluation item of the evaluation expression or a word representing the evaluation value in the character string to be processed,
In the output step, character strings extracted for each of the plurality of evaluation expressions are associated with each other and output to the related expression storage unit.
The program according to any one of claims 1 to 3, wherein: In the output step, among character strings extracted for each of the plurality of evaluation expressions, a character string including words in a common semantic class is not associated with the character strings.
The program according to claim 4. In the computer, each of words representing an evaluation item whose evaluation value can be expressed numerically among the evaluation items in the evaluation expression stored in the evaluation expression storage means, and a reference value of the evaluation value of the evaluation item, , Referring to the reference value information storage means stored in association with each other, from the character string to be processed, a word representing the evaluation item stored in the reference value information storage means, and a numerical value corresponding to the evaluation value of the evaluation item, And the evaluation expression corresponding to the combination of the evaluation item and the numerical value and its polarity based on the result of comparing the reference value associated with the extracted evaluation item and the extracted numerical value. A second specifying step for specifying in the storage means is executed;
The evaluation expression specified in the second specifying step is further processed in the classification step.
The program according to any one of claims 1 to 5, wherein: Associating an evaluation expression including a word representing an evaluation item for evaluating an evaluation object and a word representing an evaluation value for the evaluation item, and a polarity indicating whether or not the evaluation expression is a positive expression Referring to the stored evaluation expression storage means, extracting the evaluation expression included in the character string to be processed from the character string to be processed, and specifying means for specifying the polarity of each of the extracted evaluation expressions;
Classification means for classifying the extracted evaluation expressions into one or more groups based on whether the evaluation object and the polarity are common among the evaluation expressions extracted by the specifying means;
When there are a plurality of evaluation expressions classified into the same group by the classification means, an output means that associates the plurality of evaluation expressions with each other and outputs them to the related expression storage means
An information processing apparatus comprising:

Images