JP2016110440A - Term meaning learning device, term meaning determining device, method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To generate a determination rule for determining a meaning code for a term.SOLUTION: A feature-vector generation part 30 generates a feature vector on each of a plurality of technical terms. A determination rule generation part 32 generates a determination rule for determining a meaning code based on the feature vector generated for each of the plurality of technical terms and the meaning code added to each of the plurality of technical terms.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a term meaning learning device, term meaning determination device, method, and program, and more particularly, to a term meaning learning device, meaning determination device, method, and program for determining a meaning code for a term.

  Terms include notation using different characters such as hiragana, katakana, and kanji, and notation shaking such as abbreviations and abbreviations. Further, there are terms having the same meaning but different expressions such as synonyms and synonyms. For the problem of assigning a meaning code indicating the same meaning to terms of different notation or expression, whether the pair of terms is synonymous or not is described in Non-Patent Document 1. A technique for estimating using the similarity and the notation similarity can be used. Further, a technique described in Non-Patent Document 2, such as “spaghetti” and “spaghetti”, can be used to extract different notation pairs with a small edit distance that are different in some notations. Alternatively, a technique described in Non-Patent Document 3 for classifying terms based on word information used together in an actual document can be used.

Yasutomo Morimoto, Kosuke Yanai, Makoto Iwayama, “Supervised Synonym Extraction Using Context Similarity and Notation Similarity”, Proc. Of the 16th Annual Conference of the Language Processing Society, 974-977 (2010). Masahiro Kojima, Masaki Murata, Junichi Kazama, Ko Kuroda, Atsushi Fujita, Eiji Aramaki, Masaaki Tsuchida, Toshihiko Watanabe, Kentaro Torizawa, “Extracting Japanese Annotated Pairs with Small Editing Distance Using Machine Learning”, Proc. Of the 16th Annual Conference of the Language Processing Society, 928-931 (2010). Eiji Aramaki, Ken Imai, Masayuki Kanno, Kengo Miyo, Kazuhiko Oe, “Research on automatic terminology and terminology classification using medical dictionary”, Medical Informatics, Vol.25, No.6 , pp.463-474, 2006.

  The problem to be solved by the present invention is that, in the problem of assigning a pre-defined semantic code to a technical term, the most suitable semantic code for each term is selected from a large number of semantic code candidates. It is to select one.

  Unlike general words, in order to accurately convey the contents, most technical terms are considered to have a unique meaning for each specialized field. Therefore, the problem of assigning a semantic code to a terminology contained in a document with a known technical field is the same meaning but different notation or expression under the assumption that the terminology has a unique meaning. This is equivalent to giving the same meaning code to the written term.

  In the conventional technique, as described in Non-Patent Document 1 and Non-Patent Document 2, it is estimated whether two terms have the same meaning, or as in Non-Patent Document 3, using a clustering technique Dividing a set of terms into terms with similar meanings. When applying the techniques of Non-Patent Document 1 and Non-Patent Document 2 to the problem to be solved by the present invention, for example, one term example is selected for each semantic code, and the term example to which the semantic code is to be assigned Semantic codes can be estimated by examining whether they are close to

  However, in this method, there is a possibility that the estimation accuracy of the semantic code is greatly influenced by the way of selecting the term examples. On the other hand, when the technique of Non-Patent Document 3 is used for the problem to be solved by the present invention, a set of terms mechanically classified by the clustering technique and a semantic code defined manually are assigned one-to-one. It's not easy. The problem is to develop a technique for achieving high estimation accuracy for the problem of estimating the semantic code to be assigned to the technical term.

  The present invention has been made to solve the above problems, and an object of the present invention is to provide a term meaning learning device, term meaning determination device, method, and program for accurately determining the meaning code of a term. And

  In order to achieve the above object, the term meaning learning device according to the first aspect of the present invention provides a partial character included in a character string of the term for each of a plurality of terms that have been given a meaning code for the term in advance. A feature vector generation unit that generates a feature vector indicating the presence or number of each partial character string from the column, the feature vector generated for each of the plurality of terms by the feature vector generation unit, and a plurality of terms And a determination rule generation unit that generates a determination rule for determining the meaning code of the term based on the meaning code assigned to each.

  In the term meaning learning device according to the first invention, the term may be a technical term having a unique meaning defined in a specific technical field.

  Further, in the term meaning learning device according to the first invention, the feature vector generation unit, for each of the plurality of terms, from a character included in the term and a partial character string included in the character string of the term, The feature vector representing the presence or number of each character and the presence or number of each partial character string may be generated.

  Further, in the term meaning learning device according to the first invention, the feature vector generation unit, for each of the plurality of terms, from a character included in the term and a partial character string included in the character string of the term, The feature vector that represents the presence or number of each character, the presence or number of each partial character string, and the character type of each character included in the term may be generated. Further, the feature vector that further represents the character type of each partial character string used in the term may be generated.

  Further, in the term meaning learning device according to the first invention, the meaning code is based on a plurality of terms that are input with a meaning code of the term in advance and a dictionary in which synonyms or synonyms are stored. Further includes a synonym synonym meaning code assigning unit for assigning the meaning code to a term that is a synonym or similar to a term that has been assigned in advance, and the feature vector generation unit is preassigned the meaning code of the term The feature vector may be generated for each of the plurality of terms and the terms that are synonyms or similar terms to which the meaning code is assigned by the synonym synonym meaning code assignment unit.

  Further, the determination rule generation unit is configured to generate the semantic code based on the feature vector generated for each of the plurality of terms by the feature vector generation unit and the meaning code assigned to each of the plurality of terms. Each time, a binary classifier for identifying whether or not the meaning code is given to a term may be generated as the determination rule.

  A term meaning determination device according to a second invention includes a feature vector generation unit that generates a feature vector representing the presence or the number of each partial character string from a partial character string included in a character string of an input term; A semantic code determination unit that determines a semantic code of the term based on the feature vector generated by the vector generation unit and a determination rule for determining the semantic code of the term that has been learned in advance. It is configured.

  In the term meaning determination device according to the second invention, the term may be a technical term having a unique meaning defined in a specific technical field.

  Further, in the term meaning determination device according to the second invention, the feature vector generation unit includes, for the term, presence / absence of each character from a character included in the term and a partial character string included in the character string of the term. Alternatively, the feature vector representing the number and the presence / absence or number of each partial character string may be generated.

  In the term meaning learning method according to the first invention, the feature vector generation unit inputs, from each of a plurality of terms to which a term meaning code is assigned in advance, from a partial character string included in the term character string. A step of generating a feature vector representing the presence or number of each partial character string, and the determination rule generating unit generating the feature vector for each of the plurality of terms by the feature vector generating unit, and the plurality of terms Generating a determination rule for determining the meaning code of the term based on the meaning code assigned to each of the above.

  A program according to a third invention is a program for causing a computer to function as each component constituting the term meaning learning device according to the first invention or the term meaning judging device according to the second invention.

  According to the term meaning learning device, term meaning determination device, method, and program of the present invention, for each of a plurality of terms, a feature vector is generated from a character or a partial character string included in the term, and each of the plurality of terms By generating a determination rule for determining a semantic code based on the feature vector generated for and a semantic code assigned to each of a plurality of terms, it is possible to accurately determine the semantic code of a term The effect that the determination rule can be learned is obtained.

  In addition, according to the term meaning determination device, method, and program, a feature vector is generated from a character or a partial character string included in the term, and based on the generated feature vector and a previously learned determination rule, By determining the meaning code of the term, the effect that the meaning code of the term can be determined with high accuracy is obtained.

It is a block diagram which shows the functional structure of the term meaning learning apparatus which concerns on the 1st-3rd embodiment of this invention. It is a conceptual diagram which shows the index of the feature-value corresponding to the partial character string contained in a technical term. It is a block diagram which shows the functional structure of the term meaning determination apparatus which concerns on embodiment of this invention. It is a flowchart figure of the term meaning learning process routine in the term meaning learning apparatus which concerns on the 1st-3rd embodiment of this invention. It is a flowchart figure of the term meaning determination processing routine in the term meaning determination apparatus which concerns on embodiment of this invention. It is a conceptual diagram which shows the index of the feature-value corresponding to the character contained in a technical term, and a partial character string. It is a conceptual diagram which shows the index of the feature-value corresponding to the character contained in a technical term, the partial character string, and the character type of each character. It is a block diagram which shows the functional structure of the term meaning learning apparatus which concerns on the 4th Embodiment of this invention. It is a flowchart figure of the term meaning learning process routine in the term meaning learning apparatus which concerns on the 4th Embodiment of this invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

<Configuration of the term meaning learning device according to the first embodiment of the present invention>

  First, the configuration of the term meaning learning device according to the first embodiment of the present invention will be described.

  As shown in FIG. 1, the term meaning learning device 100 according to the first embodiment of the present invention stores a CPU, a RAM, a program for executing a term meaning learning processing routine described later, and various data. It can be composed of a computer including a ROM. Functionally, the term meaning learning device 100 includes an input unit 10 and a calculation unit 20 as shown in FIG.

  The input unit 10 receives, as a training data set, a plurality of technical terms to which a semantic code of technical terms has been assigned in advance, and stores it in the training data set database 28. A plurality of technical terms with semantic codes assigned in advance are technical terms used to define the semantic codes, technical terms extracted from actual documents with manual semantic codes, and synonyms And technical terms extracted from an arbitrary dictionary in which synonym relationships are described. The terminology included in any dictionary is usually not given a semantic code, so when extracting from the dictionary, the terminology used to define the semantic code and A semantic code is assigned using the relationship between synonyms and synonyms. For example, the same semantic code may be given to technical terms that are synonyms and synonyms with the technical terms used to define the semantic codes.

  The calculation unit 20 includes a training data set database 28, a feature vector generation unit 30, a determination rule generation unit 32, and a determination rule 40.

  For each of the plurality of technical terms stored in the training data set database 28, the feature vector generation unit 30 calculates a feature vector representing the number of each partial character string from the partial character strings included in the character string of the technical term. Generate. Note that a feature vector representing the presence or absence of each partial character string may be generated.

The feature vector generation unit 30 uses a partial character string constituting a technical term, for example, a character-based bi-gram, tri-gram, or the like as a feature amount, and a feature vector representing the number of each feature amount included in the technical term. Generate. For example, as shown in FIG. 2, an index of feature amounts corresponding to a partial character string included in a character string of the technical term “H. pylori” is assumed. At this time, the technical term “H. pylori” means that x _i = 1, i when i∈ {b ₁ , b ₂ , b ₃ , b ₄ , b ₅ , t ₁ , t ₂ , t ₃ , t ₄ }. Is a feature value of a partial character string not included in “H. pylori”, a feature vector ^ x = (x ₁ ,..., X _i ,..., X _I ) composed of elements with x _i = 0. Can be expressed as A symbol representing a vector is represented by adding “^” before the symbol. In FIG. 2, “_” attached to the beginning of a symbol represents the leading end of the technical term character string, and “_” attached to the end of the symbol represents the end of the technical term character string. Further, for example, a feature amount defined using a character type of a character included in a technical term such as hiragana, katakana, or alphabet may be additionally used.

  The determination rule generation unit 32 determines the meaning code of the term based on the feature vector generated for each of the plurality of technical terms by the feature vector generation unit 30 and the semantic code assigned to each of the plurality of technical terms. A determination rule is generated for

Specifically, the determination rule generation unit 32 performs, for example, a support vector machine (SVM) based on the feature vector generated for each of the plurality of technical terms and the semantic code assigned to each of the plurality of technical terms. ), Or by using a binary classifier such as a logistic regression model to estimate the value of the score function parameter ^ w _k for each semantic code k, thereby generating a determination rule mechanically. Store as 40. In the present embodiment, a technical term to which a semantic code k is assigned as a positive example and a technical term to which a semantic code other than k is assigned as a negative example with respect to the score function parameter ^ w _k of the semantic code k, Compute a parameter _{circumflex over (w) _{} k} that maximizes the objective function of the binary classifier.

<Configuration of Term Meaning Determination Device According to First Embodiment of the Present Invention>

  Next, the configuration of the term meaning determination device according to the first embodiment of the present invention will be described. As shown in FIG. 3, the term meaning determination device 200 according to the first embodiment of the present invention stores a CPU, a RAM, a program for executing a term meaning determination processing routine described later, and various data. It can be composed of a computer including a ROM. Functionally, the term meaning determination apparatus 200 includes an input unit 210, a calculation unit 220, and an output unit 250 as shown in FIG.

  The input unit 210 accepts input of terms.

  The calculation unit 220 includes a feature vector generation unit 230, a semantic code determination unit 232, and a determination rule 240.

  The determination rule 240 stores the same determination rule as the determination rule 40.

  Similar to the feature vector generation unit 30, the feature vector generation unit 230 generates a feature vector representing the number of each partial character string from the partial character string included in the term character string received by the input unit 210. Note that a feature vector representing the presence or absence of each partial character string may be generated.

  The semantic code determination unit 232 determines the semantic code of the term accepted by the input unit 210 based on the feature vector generated by the feature vector generation unit 230 and the determination rule 240.

Specifically, the semantic code determination unit 232 uses the feature vector ^ x generated by the feature vector generation unit 230 and the determination rule 240 to use the score function parameter ^ w of the determination rule 240 for each semantic code k. according to the value of _k, the score function _f k representing the degree to which the term represented by the feature vector ^ x associated with the meaning code k; calculating the _(^ x ^ _w k). Then, by selecting the semantic code k that maximizes the value of the score function f _k (^ x _m ; ^ w _k ) as the semantic code suitable for the term, the semantic code of the term is determined, and the output unit It outputs by 250. By selecting the meaning code of the term using the determination rule 240 that has been mechanically learned in advance as described above, one of the most suitable meaning codes is selected from the predefined meaning codes. Can do.

<Operation of the term meaning learning device according to the first embodiment of the present invention>

  Next, the operation of the term meaning learning device 100 according to the first embodiment of the present invention will be described. When the input unit 10 accepts a plurality of technical terms to which the meaning codes of technical terms are assigned in advance as a training data set and stores them in the training data set database 28, the term meaning learning device 100 performs the term meaning learning process shown in FIG. 4. Run the routine.

  First, in step S100, each of a plurality of technical terms stored in the training data set database 28 is acquired.

  Next, in step S102, for each of the plurality of technical terms acquired in step S100, a feature vector representing the number of each partial character string is generated from the partial character strings included in the technical term character string.

In step S104, a score is obtained for each semantic code k based on the feature vector generated for each of the plurality of technical terms in step S102 and the semantic code assigned to each of the plurality of technical terms acquired in step S100. By estimating the value of the function parameter ^ w _k , a determination rule is generated, stored in the determination rule 40, and the process ends.

  As described above, according to the term meaning learning device according to the first embodiment, for each of a plurality of technical terms, a feature vector is generated from a partial character string included in the term, and a plurality of technical terms Based on the feature vectors generated for each and the semantic codes assigned to each of the multiple technical terms, the semantic codes of the technical terms are determined with high accuracy by generating judgment rules for determining the semantic codes. It is possible to learn a determination rule for

<Operation of Term Meaning Determination Device According to First Embodiment of the Present Invention>

  Next, the operation of the term meaning determination device 200 according to the first embodiment of the present invention will be described. When the input unit 210 accepts a term, the term meaning determination device 200 executes a term meaning determination processing routine shown in FIG.

  First, in step S200, the terms accepted by the input unit 10 are acquired.

  Next, in step S202, for the term acquired in step S200, a feature vector representing the number of each partial character string is generated from the partial character string included in the character string of the technical term.

In step S204, on the basis of the feature vector generated in step S202 and the determination rule 240, the score function f _k (^ x) for each semantic code k according to the value of the parameter ^ w _k of the score function of the determination rule 240. ; ^ _{w ^} k) is calculated and the result of the calculation, the score function _f k _(^ x m; the meaning code k to the value of ^ _{w k)} is maximum, as a means code suitable for the terminology which has been acquired in step S200 By selecting, the meaning code of the term acquired in step S200 is determined.

  In step S206, the meaning code of the term obtained by the determination in step S204 is output by the output unit 250, and the process ends.

  As described above, according to the term meaning determination device according to the first embodiment, a feature vector is generated from a partial character string included in a term, the generated feature vector, and a determination rule learned in advance. By determining the meaning code of the term based on the above, the meaning code of the technical term can be determined with high accuracy.

<Configuration of Term Meaning Learning Device According to Second Embodiment of the Present Invention>

  Next, the configuration of the term meaning learning device according to the second embodiment of the present invention will be described. In addition, about the part which becomes the structure similar to the term meaning learning apparatus 100 of 1st Embodiment, the same code | symbol is attached | subjected and description is abbreviate | omitted.

  As shown in FIG. 1, the term meaning learning device 100 according to the second embodiment of the present invention stores a CPU, a RAM, a program for executing a term meaning learning processing routine described later, and various data. And a computer including a ROM. Functionally, the term meaning learning device 100 includes an input unit 10 and a calculation unit 20 as shown in FIG.

  Similar to the first embodiment, the input unit 10 according to the second embodiment accepts a plurality of technical terms to which a semantic code of technical terms is assigned in advance as a training data set, and the training data set database 28. To store.

  The computing unit 20 according to the second embodiment includes a training data set database 28, a feature vector generation unit 30, a determination rule generation unit 32, and a determination rule 40.

  The feature vector generation unit 30 according to the second embodiment includes, for each of a plurality of technical terms received by the input unit 10, a character included in the technical term and a partial character string included in the character string of the technical term Then, a feature vector representing the number of characters and the number of partial character strings is generated. Note that feature vectors representing the presence or absence of characters and the presence or absence of each partial character string may be generated.

In the feature vector generation unit 30 according to the second embodiment, characters constituting technical terms such as character-based uni-grams and partial character strings constituting technical terms such as bi-grams or tri-grams. Are used as feature quantities, and feature vectors representing the number of feature quantities included in the term are generated. For example, as shown in FIG. 6, assume an index of feature amounts of characters included in the technical term “H. pylori” and partial character strings included in the technical term character string. At this time, the technical term “H. pylori” is i∈ {u ₁ , u ₂ , u ₃ , u ₄ , b ₁ , b ₂ , b ₃ , b ₄ , b ₅ , t ₁ , t ₂ , t ₃ , x _{i =} 1, i at t _4} is a character not included in the "H. pylori", or substring of features of elements of x _{i =} 0 when the feature vector ^ x = (x ₁ , ..., x _i , ..., x _I ).

  In addition, since the other structure and effect | action of the term meaning learning apparatus which concern on 2nd Embodiment are the same as that of the term meaning learning apparatus 100 of 1st Embodiment, detailed description is abbreviate | omitted.

<Configuration of Term Meaning Determination Device According to Second Embodiment of the Present Invention>

  Next, the configuration of the term meaning determination device according to the second embodiment of the present invention will be described. In addition, about the part which becomes the structure similar to the term meaning determination apparatus 200 of 1st Embodiment, the same code | symbol is attached | subjected and description is abbreviate | omitted.

  As shown in FIG. 3, the term meaning determination apparatus 200 according to the second embodiment of the present invention stores a CPU, a RAM, a program for executing a term meaning determination processing routine described later, and various data. And a computer including a ROM. Functionally, the term meaning determination apparatus 200 includes an input unit 210, a calculation unit 220, and an output unit 250 as shown in FIG.

  The calculation unit 220 according to the second embodiment includes a feature vector generation unit 230, a semantic code determination unit 232, and a determination rule 240.

  The feature vector generation unit 230 according to the second embodiment is similar to the feature vector generation unit 30 according to the first embodiment from characters and partial character strings included in the character string of terms accepted by the input unit 210. In addition, a feature vector representing the number of characters and the number of partial character strings is generated. Note that feature vectors representing the presence or absence of each character and the presence or absence of each partial character string may be generated.

  In addition, since the other structure and effect | action of the term meaning determination apparatus based on 2nd Embodiment are the same as that of the term meaning determination apparatus 200 of 1st Embodiment, detailed description is abbreviate | omitted.

  As described above, according to the term meaning learning device according to the second embodiment, for each of a plurality of technical terms, a feature vector is generated from a character and a partial character string included in the term, and a plurality of specialized terms is obtained. Semantic codes of technical terms with high accuracy by generating judgment rules for judging semantic codes based on the feature vectors generated for each term and the semantic codes assigned to each of a plurality of technical terms Can be determined.

  Further, according to the term meaning determination device according to the second embodiment, a feature vector is generated from characters and partial character strings included in the term, and the generated feature vector and a previously learned determination rule are used. Based on the determination of the meaning code of the term, the meaning code of the technical term can be determined with high accuracy.

<Configuration of Term Meaning Learning Device According to Third Embodiment of the Present Invention>

  Next, the configuration of the term meaning learning device according to the third embodiment of the present invention will be described. In addition, about the part which becomes the structure similar to the term meaning learning apparatus 100 of 2nd Embodiment, the same code | symbol is attached | subjected and description is abbreviate | omitted.

  As shown in FIG. 1, the term meaning learning device 100 according to the third embodiment of the present invention stores a CPU, a RAM, a program for executing a term meaning learning processing routine described later, and various data. And a computer including a ROM. Functionally, the term meaning learning device 100 includes an input unit 10 and a calculation unit 20 as shown in FIG.

  The calculation unit 220 according to the third embodiment includes a training data set database 28, a feature vector generation unit 30, a determination rule generation unit 32, and a determination rule 40.

  The feature vector generation unit 30 according to the third exemplary embodiment includes a character included in the technical term and a partial character string included in the technical term for each of the plurality of technical terms received by the input unit 10. Then, a feature vector representing the number of characters, the number of partial character strings, and the character type of each character included in the technical term is generated. Note that feature vectors representing the presence or absence of each character and the presence or absence of each partial character string may be generated.

In the feature vector generation unit 30 according to the third embodiment, characters that constitute technical terms, for example, character-based uni-grams, and partial character strings that constitute technical terms, for example, bi-grams or tri-grams. The number of each feature amount included in a term that has a feature amount, etc., and the number of each feature amount included in a term that has the character type of each character that constitutes a technical term, for example, hiragana, katakana, or alphabet To generate a feature vector. For example, as shown in FIG. 7, the character included in the technical term “cervical lymphadenopathy” and the feature amount of the partial character string included in the technical term character and the character type of each character included in the technical term Assume an index combined with quantity. Here, c _t represents a t-th character in the jargon, s _t represents a character type t-th character in the terminology. If it is "cervical lymph node enlargement", for the _{c t, c 1} = neck, _{c 2} = Department, _{c 3} = can be expressed Li, and..., For _{s t is, s} 1 ₌ s 2 ₌ It can be expressed as “CJK unified Ideographs”, s ₃ = “Katakana”. Note that a feature vector representing the number of each feature quantity included in the term may be generated using the N-gram of the character type as the feature quantity.

  In addition, since the other structure and effect | action of the term meaning determination apparatus which concerns on 3rd Embodiment are the same as that of the term meaning learning apparatus 100 of 2nd Embodiment, detailed description is abbreviate | omitted.

<Configuration of Term Meaning Determination Device According to Third Embodiment of the Present Invention>

  Next, the configuration of the term meaning determination device according to the third embodiment of the present invention will be described. In addition, about the part which becomes the structure similar to the term meaning determination apparatus 200 of 2nd Embodiment, the same code | symbol is attached | subjected and description is abbreviate | omitted.

  As shown in FIG. 3, the term meaning determination device 200 according to the third embodiment of the present invention stores a CPU, a RAM, a program for executing a term meaning determination processing routine described later, and various data. And a computer including a ROM. Functionally, the term meaning determination apparatus 200 includes an input unit 210, a calculation unit 220, and an output unit 250 as shown in FIG.

  The calculation unit 220 according to the third embodiment includes a feature vector generation unit 230, a semantic code determination unit 232, and a determination rule 240.

  In addition, since the other structure and effect | action of the term meaning determination apparatus which concerns on 3rd Embodiment are the same as that of the term meaning determination apparatus 200 of 2nd Embodiment, detailed description is abbreviate | omitted.

  As described above, according to the term meaning learning device according to the third embodiment, for each of a plurality of technical terms, a feature vector is generated from the character and partial character string included in the term and the character type of the character. And by generating a determination rule for determining a semantic code based on the feature vector generated for each of the technical terms and the semantic code assigned to each of the technical terms The semantic code of the technical term can be determined.

  Moreover, according to the term meaning determination apparatus according to the third embodiment, a feature vector is generated from characters and partial character strings included in the term, and the generated feature vector and a previously learned determination rule are used. Based on the determination of the meaning code of the term, the meaning code of the technical term can be determined with high accuracy.

<Configuration of Term Meaning Learning Device According to Fourth Embodiment of the Present Invention>

  Next, the configuration of the term meaning learning device according to the fourth embodiment of the present invention will be described. In addition, about the part which becomes the structure similar to the term meaning learning apparatus 100 of 2nd Embodiment, the same code | symbol is attached | subjected and description is abbreviate | omitted.

  As shown in FIG. 8, the term meaning learning device 300 according to the fourth embodiment of the present invention stores a CPU, a RAM, a program for executing a term meaning learning processing routine described later, and various data. It can be composed of a computer including a ROM. Functionally, the term meaning learning device 100 includes an input unit 310 and a calculation unit 320 as shown in FIG.

  The input unit 310 accepts a plurality of technical terms to which a technical term meaning code is assigned in advance and a dictionary storing synonyms and synonyms of the technical terms.

  The calculation unit 320 includes a synonym synonym meaning code assigning unit 326, a training data set database 28, a feature vector generation unit 30, a determination rule generation unit 32, and a determination rule 40.

  The synonym synonym meaning code assigning unit 326 accepts the meaning code of the technical term received in the input unit 310 based on a plurality of technical terms and a dictionary, and each of the technical terms assigned the semantic code, The same meaning code is assigned to each technical term that is a synonym or synonym of the technical term. Then, the technical terms to which the semantic code is assigned in advance and the technical terms that are synonyms or synonyms to which the semantic code is assigned, which are received by the input unit 310, are stored in the training data set database 28 as a training data set.

  In addition, since the other structure and effect | action of the term meaning learning apparatus which concern on 3rd Embodiment are the same as that of the term meaning learning apparatus 100 of 2nd Embodiment, detailed description is abbreviate | omitted.

<Configuration of Term Meaning Determination Device According to Fourth Embodiment of the Present Invention>

  Next, the configuration of the term meaning determination device according to the fourth embodiment of the present invention will be described. In addition, about the part which becomes the structure similar to the term meaning determination apparatus 200 of 2nd Embodiment, the same code | symbol is attached | subjected and description is abbreviate | omitted.

  As shown in FIG. 3, the term meaning determination device 200 according to the fourth embodiment of the present invention stores a CPU, a RAM, a program for executing a term meaning determination processing routine described later, and various data. And a computer including a ROM. Functionally, the term meaning determination apparatus 200 includes an input unit 210, a calculation unit 220, and an output unit 250 as shown in FIG.

  The calculation unit 220 according to the fourth embodiment includes a feature vector generation unit 230, a semantic code determination unit 232, and a determination rule 240.

  In addition, since the other structure and effect | action of the term meaning determination apparatus which concerns on 4th Embodiment are the same as that of the term meaning determination apparatus 200 of 2nd Embodiment, detailed description is abbreviate | omitted.

<Operation of the term meaning learning device according to the fourth embodiment of the present invention>

  Next, the operation of the term meaning learning device 300 according to the fourth exemplary embodiment of the present invention will be described. When the input unit 10 accepts a dictionary storing a plurality of technical terms to which the semantic codes of technical terms are assigned in advance, and synonyms and similar terms, the term meaning learning device 300 according to the fourth embodiment The term meaning learning process routine shown in FIG. 9 is executed. In addition, about the effect | action of the term meaning learning apparatus which concerns on 4th Embodiment, about the part which becomes an effect | action similar to 2nd Embodiment, the same code | symbol is attached | subjected and description is abbreviate | omitted.

  In step S300, for each of the plurality of technical terms acquired in step S100 based on a dictionary storing synonyms and similar terms, the technical terms are synonymous with or synonymous with the technical terms. Assign the same semantic code.

  In addition, since it is the same as that of 2nd Embodiment about the other effect | action of the term meaning learning apparatus based on 4th Embodiment, detailed description is abbreviate | omitted.

  As explained above, according to the term meaning learning device according to the fourth embodiment, training with a meaning code is given by giving a meaning code to synonyms and synonyms stored in the dictionary. Data can be increased.

  Moreover, according to the term meaning determination apparatus according to the fourth embodiment, a feature vector is generated from characters and partial character strings included in the term, and the generated feature vector and a previously learned determination rule are used. Based on the determination of the meaning code of the term, the meaning code of the technical term can be determined with high accuracy.

<Experimental result>

  Distributed by MedNLP2 task (http://mednlp.jp/ntcir11/) of evaluation type workshop NTCIR-11 sponsored by National Institute of Informatics (NII) by applying the method according to the third embodiment. An experiment was conducted using a benchmark set of simulated medical documents. This benchmark set includes training data and test data, and each data includes technical terms representing disease names and symptoms to which an ICD code is assigned as a semantic code in a pseudo medical document. The ICD code is a code for pathological management and information sharing, and is a code developed for standardization of medical terms. By using the ICD code, it is possible to grasp and manage differently expressed terms having the same meaning, such as various abbreviations, kanji, and kana notation. Thousands of types of ICD codes are defined in the ICD10 compatible standard disease name master (http://www2.medis.or.jp/stdcd/byomei/).

  In the performance evaluation, when the method according to the third embodiment is applied to the problem of assigning an ICD code to a technical term representing a disease name and a symptom included in a pseudo-medical document using an ICD code as a semantic code. The determination accuracy of the obtained ICD code was examined.

  In this experiment, the training data of the benchmark set and the technical terms included in the disease name basic table included in the ICD10 compatible standard disease name master were stored in the training data set database 28. Further, technical terms included in the search table of the standard disease name master corresponding to ICD 10 are also stored in the training data set database 28. Further, the technical data synonymous with or similar to the technical terms included in the disease name basic table of the standard disease name master corresponding to the ICD10 are assigned the same ICD code as the technical terms included in the basic disease name table, and are added to the training data set database 28. Stored. In addition, life science dictionary (http://lsd.pharm.kyoto-u.ac.jp/en/index.html), T dictionary (http://www.tdic.co.jp/), and UMLS metathesaurus From each of the MedDRA dictionaries extracted from (http://www.nlm.nih.gov/research/umls/), a technical term that is synonymous with or similar to the technical term included in the ICD10-compliant standard disease name master is extracted, and The same ICD codes as the technical terms included in the ICD10-compliant standard disease name master were assigned and stored in the training data set database 28.

In this experiment, in order to generate a determination rule used in the semantic code determination unit 232, the determination rule generation unit 32 applies a logistic regression model and sets the parameter value of the score function f _k (^ x; ^ w _k ). Estimated.

  Table 1 shows the correct rate of semantic codes when a determination rule is generated by the method according to the third embodiment and applied to technical terms representing 2134 disease names and symptoms included in the test data of the benchmark set. And the number of technical terms (number of correct answers) for which the semantic code is correctly estimated. In Table 1, ADOC, MEDIS, and EXDIC represent the types of terms stored in the training data set database 28. ADIC extracted the terms stored in the training data of the benchmark set, MEDIS extracted the terms extracted from the basic disease name table and search table of the ICD10 compatible standard disease name master, and EXDIC extracted from the life science dictionary, T dictionary, and MedDRA dictionary Represents a term. It can be seen from Table 1 that the accuracy of the semantic code estimation is improved by collecting more technical terms from the dictionary and using them for generating the decision rule.

  The present invention is not limited to the above-described embodiment, and various modifications and applications can be made without departing from the gist of the present invention.

  For example, in the above-described first to fourth embodiments, a term meaning learning device that generates a determination rule for determining a meaning code from a technical term having a unique meaning defined in a specific specialized field is taken as an example. As described above, the present invention is not limited to this, and if a term with a semantic code is used in advance, a determination rule can be generated and a semantic code can be determined for terms in any field. it can. For example, the term meaning learning device generates a determination rule for a term in a certain field (natural science, economy, etc.), and the term meaning determination device uses the generated determination rule to generate a term in the field. The semantic code can be determined.

  Further, the case where the term meaning learning device and the term meaning determination device are configured as separate devices has been described as an example, but the present invention is not limited to this, and the term meaning learning device and the term meaning determination device are one device. You may make it comprise.

10, 210, 310 Input unit 20, 220, 320 Operation unit 28 Training data set database 30, 230 Feature vector generation unit 32 Determination rule generation unit 40, 240 Determination rule 100, 300 Term meaning learning device 200 Term meaning determination device 232 Meaning Code determination unit 250 Output unit 326 Synonym synonym meaning code giving unit

Claims (8)

A feature vector for generating a feature vector representing the presence or number of each partial character string from the partial character strings included in the character string of each term, for each of a plurality of terms to which the meaning code of the term is given in advance A generator,
A determination rule for determining a semantic code of a term based on the feature vector generated for each of the plurality of terms by the feature vector generation unit and a semantic code assigned to each of the plurality of terms. A determination rule generation unit to generate,
A term meaning learning device.   The term meaning learning device according to claim 1, wherein the term is a technical term whose unique meaning is defined in a specific technical field.   The feature vector generation unit, for each of the plurality of terms, from the character included in the term and the partial character string included in the character string of the term, the presence or absence of each character and the presence or absence of each partial character string The term meaning learning device according to claim 1, wherein the feature vector representing the number or the number is generated. A feature vector generation unit that generates a feature vector representing the presence or number of each partial character string from the partial character string included in the character string of the input term;
A semantic code determination unit that determines a semantic code of the term based on the feature vector generated by the feature vector generation unit and a determination rule for determining a semantic code of the term that has been learned in advance;
A term meaning determination device including:   The term meaning determination device according to claim 4, wherein the term is a technical term whose unique meaning is defined in a specific technical field.   The feature vector generation unit, for the term, from the character included in the term and the partial character string included in the character string of the term, the presence or number of each character and the presence or number of each partial character string The term meaning determination device according to claim 4 or 5, wherein the feature vector to be expressed is generated. The feature vector generation unit represents the presence or number of each partial character string from the partial character strings included in the character string of the term for each of the plurality of terms to which the meaning code of the term is input in advance Generating a vector;
A determination rule generation unit determines a semantic code of a term based on the feature vector generated for each of the plurality of terms by the feature vector generation unit and a semantic code assigned to each of the plurality of terms Generating a determination rule for
A term meaning learning method.   A computer is made to function as each part which comprises the term meaning learning apparatus of any one of Claims 1-3, or the term meaning determination apparatus of any one of Claims 4-6. Program for.