JP2008097286A - Apparatus for retrieving opinion sentence, method for retrieving opinion sentence, program for retrieving opinion sentence, and recording medium recording the program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an apparatus for retrieving opinion sentence capable of outputting a retrieval result corresponding to a degree of fine opinion likeness independently of a field of a retrieval word or a format of a document. <P>SOLUTION: A retrieval result sentence acquisition part 12 acquires a sentence in which a retrieval word input from a retrieval word input part 11 appears from a sentence storage part 15 and an opinion likeness index attachment part 17 calculates an opinion contribution degree indicating which degree an opinion-related expression in a sentence exerts influence on judgment of the sentence as an opinion from an index of opinion likeness, attaches the opinion contribution degree to the opinion-related expression, calculates and attaches an index of opinion likeness in each sentence on the basis of the opinion contribution degree, and stores all sentences to which the index of opinion likeness is attached in an opinion likeness index attached sentence storage part 16. An opinion sentence acquisition part 13 acquires an opinion sentence with the opinion likeness index and an opinion sentence output part 14 outputs the acquired opinion sentence on the basis of the opinion likeness index. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to an opinion sentence search device, an opinion sentence search method and an opinion sentence search program that output an opinion sentence from a plurality of electronic sentences in which opinions are described, and a recording medium on which the program is recorded.

  There is a known method for retrieving opinion information from documents containing opinions in natural language existing on the Internet or questionnaire surveys. For example, in Patent Document 1 below, a survey result of a questionnaire (for example, a questionnaire on road administration) limited to an opinion object is used to determine whether the opinion is an opinion using an analysis pattern prepared in advance, and an opinion sentence is output. is doing.

  Further, in Patent Document 2 below, it is a method of searching reputation information (information including an expression that can determine the value of an object) instead of an opinion. Whether or not the sentence is satisfied for each rule regarding reputation. For each combination of values indicating such, a score of reputation is given using a score input by a person, and search results are output in the order of reputation.

As a method for collecting opinion sentences, for example, in Patent Document 3 below, an accurate opinion is extracted by determining the likelihood of opinion based on syntactic aptitude or objective aptitude.
JP 2004-227343 A JP 2002-175330 A JP 2003-203136 A

  There are various forms of documents that describe opinions in a wide range of fields, including questionnaire survey results, reviews of movies and products, impressions of events attended by individuals, personal views of tourist spots and people, and individual claims regarding events. There are many documents.

  However, in the above-mentioned Patent Document 1, since retrieval is performed from a document in which an opinion target field is limited in advance, it is possible to give an opinion-likeness in that field by creating an analysis pattern including an opinion-like index. There is a problem that it is not realistic to comprehensively prepare analysis patterns corresponding to all documents of various formats in a wide range of fields.

  Further, in Patent Document 2, it is possible to make a rule regarding reputation likeness as a rule for opinion likeness, and if it is an amount of rules that can be created manually, a score of opinion likeness can be given manually. There is a high possibility that the same opinion-likeness will be obtained with a small amount of wide rules, and there is a problem that sentences with the same opinion-likeness cannot be displayed in the order of appropriate opinion-likeness. Giving opinion-likeness to a large number of individual rule combinations reduces the possibility that the target sentence will have the same opinion-like score, and solves this problem. It is not realistic to manually give an opinion score.

  Further, in Patent Document 3, the likelihood of opinion is determined by syntactic aptitude or objective characteristics, and the syntactic characteristics are determined by a small number of rules that do not depend on the target field or form. There is a high possibility that the scores of the same opinion are likely to be obtained, and there is a problem that the sentences of the same opinion cannot be displayed in the order of appropriate opinions. As for objective characteristics, there is a problem that it is not possible to rank the likelihood of opinion only by objective characteristics because it only determines whether or not the target sentence is described by a third party.

  The present invention has been made in view of the above problems, and an opinion sentence search device capable of outputting a search result according to the degree of fineness of opinion regardless of the field of the search word or the format of the document, An object is to provide an opinion sentence search method, an opinion sentence search program, and a recording medium in which the program is recorded.

  Specifically, from the data in which the result of manual opinionality determination (determination of whether or not it is an opinion) is given in advance, for each expression (or all expressions) related to the opinion, Appropriate contribution to the opinion according to the data without using manpower by calculating the degree of opinion contribution according to the judgment result of the opinion (degree that contributes to judging the opinion) and assigning the value A degree is given to each expression, and for a sentence of a document that includes the input search term, an arbitrary likelihood is calculated by calculating the likelihood of the sentence based on the degree of opinion contribution given to each expression. Opinion sentence search results are displayed in such a way that the likelihood of opinion is calculated and the opinion likelihood is ranked more finely and appropriately according to the result, without manually creating a huge number of word combination rules for the field of opinion. It is possible Opinion text search apparatus, there is provided a recording medium recording the opinion text search method and sentiments search program and a program.

  The opinion sentence search device according to claim 1 for solving the above-mentioned problem is a device for outputting an opinion sentence related to a search word from a plurality of sentences, a search word input means for inputting a search word, and the search word Search result sentence acquisition means for acquiring a sentence in which a search word input by the input means appears from a sentence storage device storing a plurality of sentences, and opinions for each sentence with respect to all sentences in the sentence storage device A search result sentence acquired by the search result sentence acquisition means, and an opinion likelihood index adding means for storing in the sentence storage device with an opinion likelihood index all the sentences to which the likelihood index is assigned To obtain an opinion sentence to which an opinion likelihood index is assigned based on an opinion likelihood index necessary to determine the degree of the opinion sentence stored in the sentence storage device with the opinion likelihood index Acquisition means and The has been sentiments acquired by the opinion sentence obtaining unit is characterized by having an opinion sentence output means for outputting an opinion statement based on the indication of the opinion ness.

  Further, the opinion sentence search device according to claim 2 is the opinion sentence search device according to claim 1, wherein the opinion likelihood index assigning unit is configured to 1 from an opinion related expression storage device storing expressions related to opinions. An opinion related expression acquiring means for acquiring one opinion related expression, a sentence including one opinion related expression acquired by the opinion related expression acquiring means, and an index of the likelihood of opinion given to the sentence. Which of the opinion-related expressions in the sentence is used to determine that the sentence is an opinion from the opinion-like index acquisition means acquired from the indexed sentence storage device and the opinion-like index acquired by the opinion-likeness index acquisition means An opinion contribution degree calculating means for calculating an opinion contribution degree indicating a degree of influence, and storing the opinion related expression assigned the opinion contribution degree in the opinion related expression storage device; Opinion related expression end determination means for checking whether or not there is an unprocessed opinion related expression stored in the opinion related expression storage device, and the opinion related expression end determination means determines the end. A sentence acquisition unit that acquires a full sentence from the sentence storage device, and an opinion related expression that appears in one sentence stored in the opinion related expression storage device with respect to the sentence acquired by the sentence acquisition unit. It includes an opinion-likeness index calculation means for calculating the opinion-likeness of the one sentence from the degree of opinion contribution and storing the sentence and the likeness-likeness in a sentence storage device with an opinion-likeness index.

  The opinion sentence search method according to claim 3 is a method for outputting an opinion sentence related to a search word from a plurality of sentences, a search word input step in which the search word input means inputs the search word, and a search result sentence acquisition A search result sentence acquisition step for acquiring a sentence in which the search word input in the search word input step appears from a sentence storage device in which a plurality of sentences are stored, and an opinionality index assigning means, Opinion-likeness index giving step for assigning an opinion-like index for every sentence to the whole sentence of the sentence storage device, and storing all sentences given the opinion-likeness index in the sentence storage device with the opinion-likeness index, Opinion necessary for the sentence acquisition means to determine the degree of the opinion sentence stored in the sentence storage device with the likelihood index from the search result sentence acquired by the search result sentence acquisition step An opinion sentence acquisition step for acquiring an opinion sentence to which an opinion likeness index has been assigned based on the index of likelihood, and an opinion sentence output means that converts the opinion sentence acquired by the opinion sentence acquiring step into the opinion likelihood index And an opinion sentence output step for outputting an opinion sentence based on the above.

  The opinion sentence search method according to claim 4 is the opinion sentence search method according to claim 3, wherein the opinion-related index assigning step stores an expression related to the opinion. An opinion-related expression acquisition step for acquiring one opinion-related expression from the opinion-related expression storage device, and a sentence including the one opinion-related expression acquired by the opinion-related expression acquisition means by the opinion-related expression acquisition step and the sentence An opinion-likeness index acquisition step for acquiring an opinion-likeness index assigned to the opinion-likeness-indexed sentence storage device, and an opinion contribution degree calculating means for the opinion-likeness index acquired by the opinion-likeness index acquisition step. Based on the index, an opinion contribution degree indicating how much the opinion-related expression in the sentence has an influence on judging the sentence as an opinion is calculated. The opinion contribution calculation step of storing the opinion related expression to which the opinion contribution is given in the opinion related expression storage device, and the opinion related expression end determination means are stored in the opinion related expression storage device, the opinion contribution An opinion-related expression end determination step for checking whether or not there is an unprocessed opinion-related expression, and a sentence acquisition unit acquires a full sentence from the sentence storage device when it is determined to be ended by the opinion-related expression end determination step A sentence acquisition step, and an opinion-likeness index calculation means, with respect to the sentence acquired by the sentence acquisition step, an opinion contribution of an opinion related expression appearing in one sentence stored in the opinion related expression storage device Calculating an opinion-likeness of the sentence, and storing the sentence-likeness in the sentence storage device with the opinion-likeness index; It is characterized in that it comprises.

  In the above configuration, the opinion-like index given by the opinion-likeness index assigning means (opinion-likeness index assigning step) is an index indicating whether the inputted sentence is a sentence expressing an opinion, and the sentence is an individual It is the degree to which it can be judged that it is an opinion sentence such as the principle, assertion, opinion, impression, and evaluation.

  For example, the value (indicator) of the likelihood of opinion is set in three stages: when it can be determined that it is not an opinion, when neither can be determined, and when it can be determined that it is an opinion.

  Further, the opinion-likeness index giving means (an opinion-likeness index giving step) assigns an opinion contribution to the opinion-related expression, and calculates the opinion-likeness using the opinion contribution.

  For this reason, since the index of opinion-likeness is calculated in detail by the combination of opinion contributions given for each opinion-related expression that appears in the sentence, the evaluation granularity of the opinion-likeness of the sentence (how much the index of opinion-likeness is expressed (For example, the granularity is finer than that of the three-stage index), and accordingly, the likelihood of opinion can be appropriately ordered.

  An opinion sentence search program according to claim 5 is a program for causing a computer to execute the opinion sentence search method according to claim 3 or 4.

  A recording medium according to claim 6 is characterized in that the program according to claim 5 is recorded on a recording medium readable by the computer.

(1) According to the inventions described in claims 1 to 6, in the opinion sentence search in which a search word is input and an opinion sentence is output from a document, the likelihood of opinion (sentence) regardless of the field of the search word or the form of the document It is possible to output a search result according to the degree of how much it can be determined that is an opinion sentence.
(2) According to the second and fourth aspects of the invention, the evaluation granularity of the likelihood of opinion (how much it can be determined that the sentence is an opinion sentence) regardless of the field of the search term or the format of the document ( The degree to which the degree of opinion can be expressed in detail) can be made finer, and the search results corresponding to the degree of opinion-likeness can be output more finely.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings, but the present invention is not limited to the following embodiments. First, the principle of the present invention will be described with reference to FIGS. FIG. 1 is a flowchart for explaining the present invention.

  In FIG. 1, the present invention inputs a search word (step S1), acquires a sentence in which the input search word appears, and obtains a search result sentence from a sentence storage device in which a plurality of sentences are stored (step S1). S2) An opinion-like index is assigned based on the opinion-like index necessary for judging the degree of the opinion sentence stored in the sentence storage device with the opinion-like index from the acquired search result sentence. The obtained opinion sentence is acquired (step S3), and an opinion sentence set is output based on the acquired opinion likelihood index of the opinion sentence with the index of opinion likelihood (step S4).

  The opinion-like index is an index indicating whether the input sentence is a sentence expressing an opinion, and to what extent the sentence is an opinion sentence such as an individual's principle, claim, opinion, impression, evaluation, etc. It is the degree of judgment.

  FIG. 2 shows a principle configuration diagram of the present invention.

  In FIG. 2, the present invention provides a search word input means 1 for inputting a search word, and a sentence storage device in which a plurality of sentences are stored as a sentence in which the search word input by the search word input means 1 appears. Search result sentence acquisition means 2 acquired from 5 and the sentence-likeness index for every sentence given to the whole sentence of the sentence storage device 5, and all sentences with the opinion-likeness index are stored as sentence-like sentences with opinion-likeness index The degree of opinion sentence stored in the sentence storage apparatus 6 with opinion-likeness index is determined from the opinion-likeness index assigning means 7 stored in the device 6 and the search result sentence acquired by the search result sentence acquiring means 2. An opinion sentence acquisition means 3 for obtaining an opinion sentence to which an opinion likeness index is given based on an index of opinion likeness necessary for the evaluation, and an opinion sentence acquired by the opinion sentence acquiring means 3 as an index of opinion likelihood Output opinion text based on And an opinion sentence output means 4.

  FIG. 3 is a block diagram showing an opinion sentence search apparatus according to an embodiment of the present invention. The opinion sentence search device includes a search word input unit 11, a search result sentence acquisition unit 12, an opinion sentence acquisition unit 13, an opinion sentence output unit 14, a sentence storage unit 15, a sentence storage unit 16 with an opinion-likeness index, and an opinion-likeness index assignment unit 17.

  The search word input unit 11 inputs a search word.

  The search result sentence acquisition unit 12 acquires a sentence in which the search word input by the search word input unit 11 appears from the sentence storage unit 15 in which a plurality of digitized sentences are stored.

  The opinion-likeness index assigning unit 17 assigns an opinion-likeness index to the entire sentence in the sentence storage unit 15 for each sentence, and stores all sentences to which the opinion-likeness index is assigned in the sentence-likeness-attached sentence storage unit 16. To do.

  The opinion sentence acquisition unit 13 uses the search result sentence acquired by the search result sentence acquisition unit 12 to determine the degree of opinion necessary to determine the degree of the opinion sentence stored in the opinion storage index-added sentence storage unit 16. Based on the index, an opinion sentence with an opinion-like index is obtained.

  The opinion sentence output unit 14 outputs the opinion sentence acquired by the opinion sentence acquisition unit 13 based on the index of opinion likelihood.

  Search word input unit 11, search result sentence acquisition unit 12, opinion sentence acquisition unit 13, opinion sentence output unit 14, sentence storage unit 15, sentence storage unit 16 with opinion-likeness index and opinion-likeness constituting the opinion sentence search device. Each function of the index providing unit 17 is executed by a computer, for example.

  Next, the operation of the opinion sentence search apparatus will be described.

  FIG. 4 is a flowchart showing opinion sentence search processing according to an embodiment of the present invention. Hereinafter, an example of an opinion sentence search that outputs an opinion sentence when the search word “car B” is input will be described based on this flowchart.

  In step S <b> 11, a search word is input by the search word input unit 11. In this example, “car B” is input.

  In step S <b> 12, the search result sentence acquisition unit 12 acquires a document including a search word for each sentence stored in the sentence storage unit 15. For example, if the document shown in FIG. 10 is stored in the sentence storage unit 15, the document is acquired because it is a document including the search term “car B”.

  In step S <b> 13, the opinion sentence acquisition unit 13 uses a sentence storage unit with an opinion-likeness index as a sentence to which a predetermined number or more of opinion-likeness is assigned among the plurality of sentences extracted by the search result sentence acquisition unit 12. 16 from. The sentence with an opinion-like index used here may be given an opinion-likeness to all sentences in the sentence storage unit 15 by the assigning part 17 after the step S13. An example of this will be described with reference to the drawings.

  FIG. 5 is a configuration diagram of opinion-likeness imparting means according to an embodiment of the present invention.

  Opinion-likeness index giving means 7 (opinion-likeness index giving unit 17) includes opinion-related expression acquisition unit 71, opinion-likeness index acquisition unit 72, opinion contribution degree calculation unit 73, opinion-related expression end determination unit 74, and sentence acquisition unit 75. , Opinion-likeness index calculation unit 76, opinion-related expression storage unit 77 as an opinion-related expression storage device, learning-like opinion-like sentence-attached sentence storage unit 78, opinion-related expression storage device with opinion contribution as an opinion-related expression storage device The unit 79, the sentence storage unit 15, and the sentence storage unit 16 with opinion-likeness index are configured.

  The opinion related expression acquisition unit 71 acquires one opinion related expression stored in the opinion related expression storage unit 77. Here, the opinion-related expression is an expression related to an opinion necessary to be judged as an opinion, and is manually extracted from a document in which an opinion is described in the opinion-related expression storage unit 77. Keep it. When it is difficult to define and extract opinion-related expressions, all words appearing in a document in which opinions are described may be used.

  The opinion-likeness index acquisition unit 72 acquires a sentence including the opinion-related expression acquired by the opinion-related expression acquisition unit 71 and the opinion-likeness index from the learning opinion-likeness indexed sentence storage unit 78. The sentence with the opinion-likeness index for learning is data to which a person has an opinion about a document in which an opinion is described.

  The opinion contribution degree calculation unit 73 calculates the opinion contribution degree of the opinion-related expression from the opinion-like index acquired by the opinion-like index acquisition unit 72, and contributes the opinion-related expression to which the opinion contribution degree is given to the opinion contribution It is stored in the opinion related expression storage unit 79 with degree. Here, the opinion contribution is an index indicating how much the opinion related expression contributes to the judgment of opinion.

  In the present embodiment, the opinion related expression storage unit 79 with the opinion contribution degree stores the same data (opinion related expressions) as the opinion related expressions stored in the opinion related expression storage unit 77 in advance. is there.

  The opinion related expression end determination unit 74 checks whether or not there is an unprocessed opinion related expression stored in the opinion related expression storage unit 79 with opinion contribution.

  The sentence acquisition unit 75 acquires all the sentences stored in the sentence storage unit 15.

  The opinion-likeness index calculation unit 76 has an opinion property of an opinion-related expression appearing in one sentence stored in the opinion-related expression storage unit 79 with an opinion contribution for each sentence acquired by the sentence acquisition unit 75. Calculate the likelihood of the sentence from the contribution. In addition, the calculated opinion-likeness index is stored in the opinion-likeness-indexed sentence storage unit 16 together with the sentence.

  FIG. 6 is a flowchart for giving an opinion likeness according to an embodiment of the present invention. Hereinafter, based on this flowchart, a method of giving an opinion-likeness index will be described.

  In step S <b> 71, the opinion related expression acquisition unit 71 acquires one opinion related expression stored in the opinion related expression storage unit 77. For example, when the opinion related expression shown in FIG. 7 is stored in the opinion related expression storage unit 77, first, “high” is extracted.

  In step S72, the opinion-likeness index acquisition unit 72 acquires all the sentences including the opinion-related expressions acquired in step S71 and the opinion-likeness index from the learning-like opinion-like sentence-attached sentence storage unit 78. In the sentence storage section 78 with an opinion-likeness index for learning, for example, a sentence and an opinion-like index shown in FIG. 8 are stored.

  For example, the value (indicator) of the likelihood of opinion is set to the following three stages: 1 when it can be judged that it is not an opinion, 2 when neither can be judged, and 3 when it can be judged as an opinion At that time, “100 yen high” in FIG. 8 is not an opinion, so the likelihood of opinion is 1, “Bill is high”, because neither could be judged, the likelihood of opinion is 2, “Pride is high” Since it was judged as an opinion, the likelihood of opinion is 3.

  If you get a sentence that contains “high” and an index of opinion from FIG. 8, it is “100 yen high”, the opinion is “1”, “the price is high”, the opinion is “2”, and “the building is expensive” “2” and “3” are obtained that are “pride is high”. Note that this opinion-likeness is given in advance by a person to the sentence in advance. In the present embodiment, a numerical value “1”, “2” or “3” is given. “1” was determined to be not an opinion, “3” was determined to be an opinion, and “2” was determined to be neither.

  The index is not limited to the present embodiment as long as it is a numerical value that can be given to a person as an opinion. For example, the index may be any value such as five levels. Further, it is not necessary to use only the data given by one person, and any number of people who give a feeling of opinion, such as taking an average value of data given by a plurality of people, may be used.

  In step S73, the opinion contribution calculation unit 73 calculates the opinion contribution degree of the opinion related expression from the opinion likelihood index acquired by the opinion likelihood index acquisition unit 72, and assigns the opinion contribution degree. The expression is stored in the opinion related expression storage unit 79 with opinion contribution.

  For example, if the opinion contribution is defined as a numerical value that is normalized so that the value is 0 to 1 with respect to the overall average, if “high”, the average is (1 + 2 + 2 + 3) / 4 = 2, and thus normalization is performed. 2/3 = 0.67. There are other methods for calculating the opinion contribution, such as weighting according to the total value of opinion-like values and the overall appearance distribution of opinion-related expressions. Any index can be used as long as it is an indicator that can be determined.

  In steps S74a and S74b, the opinion related expression end determination unit 74 checks whether or not there is an unprocessed opinion related expression stored in the related expression storage unit 79 with opinion contribution. In the present example, since unprocessed opinion-related expressions remain, it is determined to be “present”, and unprocessed opinion-related expressions stored in the opinion-related expression storage unit 77 are acquired. From FIG. 7, the next unprocessed opinion-related expression is “recommended”.

  In this way, by repeating steps S71 to S74b, for example, an opinion contribution degree is given to each opinion related expression shown in FIG. When there is no unprocessed opinion-related expression, it is determined that there is no expression, and the process proceeds to the next step S75.

  In step S75, the sentence acquisition unit 75 acquires all the sentences stored in the sentence storage unit 15. For example, when the sentences shown in FIG. 10 are stored, all of these sentences are acquired.

  In step S76, the opinion-likeness index calculation unit 76 generates an opinion related expression that appears in the sentence stored in the opinion related expression storage unit 79 with opinion contribution for each sentence acquired by the sentence acquisition unit 75. Calculate the likelihood of the sentence from the opinion contribution. In addition, the calculated opinion-likeness index is stored in the opinion-likeness-indexed sentence storage unit 16 together with the sentence.

  This “likeness of opinion” is an index indicating whether or not the input sentence is a sentence expressing an opinion, and is defined as, for example, an average value of the degree of opinion contribution assigned to the appearing opinion-related expression. Thus, for example, the sentence ID 1 of FIG. 10 “Tested a car B by company A” does not include the opinion-related expression of FIG. In addition, sentence ID2 “Vehicle height is high” in FIG. 10 includes the opinion related expression “high” in FIG. 9, so it becomes 0.67, and sentence ID3 “Recommended for large people.” Includes the opinion-related expressions “large”, “definitely”, and “recommended” in FIG. 9, so that (0.67 + 1 + 1) /3=0.89. Similarly, the sentence ID 4 “I think the price is very high” includes the opinion related expressions “Great”, “High”, and “I think” in FIG. 9, so (1 + 0.67 + 0. 9) /3=0.86, and the results shown in FIG. 11 are obtained for all sentences.

  As described above, in the present embodiment, real values of 0 to 1 as shown in FIG. 11 are used as indicators of opinion likeness. Therefore, the granularity (3, 1, and 3) shown in FIG. The evaluation granularity of opinion-likeness indicating how finely the opinion-likeness index can be expressed is fine.

  If the index of opinion-likeness is an index that can be used to understand the opinion-likeness of the sentence using the opinion contributions assigned to the opinion-related expressions, for example, the total value of opinion contributions and the appearance frequency of opinion-related expressions The calculation method is not limited to this example, such as weighting opinion contributions accordingly.

  An opinion sentence (in this example, a sentence whose opinion likelihood is not 0) to which an opinion likelihood index is assigned is stored in the opinion storage index-added sentence storage device 16 using the method as in the above example.

  In the embodiment of FIG. 5, the opinion related expression storage device of the present invention is composed of two storage units, an opinion related expression storage unit 77 and an opinion related expression storage unit 79 (rewritable memory) with opinion contribution. However, the present invention is not limited to this, and a single rewritable opinion-related expression storage unit may be used.

  When configured in this way, the opinion related expression acquisition process of the opinion related expression acquisition unit 71 (step S71 in FIG. 6), the process of storing the opinion related expression with the opinion contribution calculated by the opinion contribution degree calculation unit 73 (FIG. 6 in step S73), the opinion-related expression end determination unit 74 checks whether or not there is an unprocessed opinion-related expression (steps S74a and S74b in FIG. 6), and the opinion-likeness index calculation unit 76 calculates the likelihood of opinion. The process (step S76 in FIG. 6) is performed on the one opinion related expression storage unit.

  Next, the process after step S13 is demonstrated using FIG.

  In step S14, the sentences with the opinion-likeness acquired in step S13 are output in descending order of opinion-likeness. Since the example sentence of FIG. 10 is given the opinion-likeness of FIG. 11 (the same sentence ID of FIG. 10 and the sentence ID of FIG. 11 indicate the same sentence), the opinion sentence (in this example) If sentences with opinion-likeness other than 0) are output in descending order of opinion-likeness, they are displayed in the order shown in FIG. In this example, it was output in order of high opinion opinion, but for example, if you set a high threshold of opinion likeness and set to output only opinion sentences that seem more opinion, or if there are many opinion sentences, Any output method that uses opinion-like values, such as a method of limiting the number of opinion sentences to be output, may be used.

  When calculating the opinion likeness using the opinion contribution as described above, the opinion likeness is calculated in detail by the combination of the opinion contribution given for each expression that appears in the sentence, so the evaluation granularity of the opinion likeness of the sentence It becomes fine. Thereby, the case where it becomes the same point (same value) decreases, and it is possible to order opinions more appropriately.

  That is, the opinion contribution calculation unit 73 in FIG. 5 can give an opinion contribution for each expression, and the opinion likelihood index calculation unit 76 combines the opinion contributions given to each expression that appears in the sentence. Thus, an opinion-like index is calculated, so that the evaluation granularity of opinion-likeness can be automatically made finer for any sentence as compared with the conventional method. As a result, even if the same expression appears, the value of opinion changes depending on the combination with other expressions.

  For example, even when the same “high” appears, the granularity of the index of opinion can be reduced from the appearance status of other words. In the example of FIG. 10, “high” appears in both of the sentence ID 4 “the price is very high” and the sentence ID 2 “the vehicle height is high”, as shown in FIG. 9. Sentence contributions are also given to “Great” and “I think”, so the sentence-like values calculated using the opinion contributions are different, and the sentence can be judged to be more opinion The sentence with ID4 is ranked higher than the sentence ID2. Thereby, this invention can rank and display the likelihood of opinion more appropriately.

  In addition, the said Example can be grasped | ascertained as invention of a program.

  That is, in the above embodiment, in a program that outputs an opinion sentence related to a search word from a plurality of sentences, a search word input step for inputting the search word and a search word input by the search word input step appear. A search result sentence acquisition step of acquiring a sentence from a sentence storage device storing a plurality of sentences, and assigning an opinion-like index for each sentence to the whole sentence of the sentence storage device, From the retrieval result sentence obtained by the opinion-likeness index adding step for storing all the given sentences in the opinion-likeness-indexed sentence storage device and the search result sentence obtaining step, the sentence is stored in the sentence-likeness index-added sentence storage device. An opinion sentence obtaining step for obtaining an opinion sentence to which an opinion-like index is assigned based on an opinion-like index necessary for judging the degree of the opinion sentence; and A is opinion sentence obtaining by Mibun acquiring step is an example of a program for executing an opinion sentence output step of outputting an opinion statement based on the indication of the opinion likeness to the computer.

  The opinion-likeness index assigning step is acquired by an opinion-related expression acquisition step of acquiring one opinion-related expression from an opinion-related expression storage device storing expressions related to an opinion, and the opinion-related expression acquisition step. An opinion likelihood index acquisition step of acquiring a sentence including one opinion related expression and an opinion likelihood index assigned to the sentence from a sentence storage device with an opinion likelihood index, and the opinion likelihood index acquisition step. The opinion contribution that indicates the degree to which the opinion-related expression in the sentence has an influence on the judgment of the sentence as an opinion is calculated from the index of the likelihood of opinion, and the opinion that has been given the opinion contribution An opinion contribution degree calculating step of storing a related expression in the opinion related expression storage device, and the meaning stored in the opinion related expression storage device Opinion related expression end determination step for examining presence / absence of unprocessed opinion related expression in the contribution degree giving process, and sentence acquisition step for acquiring a full sentence from the sentence storage device when it is determined to be ended by the opinion related expression end determination step And, for the sentence acquired by the sentence acquisition step, the opinion likeness of the sentence is calculated from the opinion contribution degree of the opinion related expression appearing in the sentence stored in the opinion related expression storage device, It is an example of a program to be executed by a computer, including an opinioniness index calculation step of storing the sentence and the likelihood of opinion in a sentence storage device with an opinioniness index.

  It is also possible to supply a recording medium recording the program to a system or apparatus, and the CPU (MPU) of the system or apparatus reads and executes the program stored in the recording medium. In this case, the program itself read from the recording medium realizes the functions of the above-described embodiment, and examples of the recording medium on which the program is recorded include CD-ROM, DVD-ROM, CD-R, CD- There are RW, MO, and HDD.

  The present invention is not limited to the above-described embodiments and examples, and various modifications and applications can be made within the scope of the claims.

It is a flowchart for demonstrating the principle of this invention. It is a principle block diagram of this invention. It is a basic block block diagram of the opinion sentence search apparatus which is one Example of this invention. It is a flowchart of the opinion sentence search process which is one Example of this invention. It is a block diagram of the opinion likeness provision means which is one Example of this invention. It is a flowchart of opinion-like provision which is one Example of this invention. It is explanatory drawing which shows an example of the opinion related expression which is one Example of this invention. It is explanatory drawing which shows an example of the sentence with the index of the opinion-likeness for learning which is one Example of this invention. It is explanatory drawing which shows an example of the opinion related expression with an opinion contribution which is one Example of this invention. It is explanatory drawing which shows an example of the document which is one Example of this invention. It is explanatory drawing which shows an example of the opinion likeness index | index provided for every sentence which is one Example of this invention. It is explanatory drawing which shows the example of an opinion sentence output which is one Example of this invention.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... Search word input means, 2 ... Search result sentence acquisition means, 3 ... Opinion sentence acquisition means, 4 ... Opinion sentence output means, 5 ... Sentence storage apparatus, 6 ... Sentence storage apparatus with an opinion-likeness index, 7 ... Opinion-likeness index Giving means: 11 ... Search word input unit, 12 ... Search result sentence acquisition unit, 13 ... Opinion sentence acquisition part, 14 ... Opinion sentence output part, 15 ... Sentence storage part, 16 ... Sentence storage part with opinion-likeness index, 17 ... Opinion-likeness index assigning unit, 71 ... Opinion-related expression acquisition unit, 72 ... Opinion-likeness index acquisition unit, 73 ... Opinion contribution degree acquisition unit, 74 ... Opinion contribution degree end determination unit, 75 ... Sentence acquisition unit, 76 ... Opinion likeness Index calculating unit, 77... Opinion related expression storage unit, 78... Sentence storage unit with an opinion-like index for learning, and 79.

Claims (6)

In a device that outputs an opinion sentence related to a search word from a plurality of sentences,
A search term input means for inputting a search term;
Search result sentence acquisition means for acquiring a sentence in which a search word input by the search word input means appears from a sentence storage device storing a plurality of sentences;
An opinion-likeness index giving means for assigning an opinion-like index for each sentence to the whole sentence of the sentence storage device, and storing all sentences given the opinion-likeness index in the sentence storage device with the opinion-likeness index;
Based on the opinion-like index necessary for judging the degree of the opinion sentence stored in the sentence storage device with the opinion-like index from the sentence of the search result obtained by the search result sentence obtaining means Opinion sentence acquisition means for acquiring an opinion sentence with an index of uniqueness,
An opinion sentence search device, comprising: an opinion sentence output means for outputting an opinion sentence obtained by the opinion sentence acquisition means based on the opinion likelihood index. The opinion-likeness index giving means is:
An opinion related expression acquisition means for acquiring one opinion related expression from an opinion related expression storage device in which an expression related to the opinion is stored;
Opinion-like index acquisition means for acquiring a sentence including one opinion-related expression acquired by the opinion-related expression acquiring means and an opinion-like index assigned to the sentence from the sentence storage device with an opinion-like index for learning When,
From the opinion-like index acquired by the opinion-likeness index acquisition means, an opinion contribution degree indicating the degree to which the opinion-related expression in the sentence affects the judgment of the sentence as an opinion is calculated. , An opinion contribution degree calculating means for storing the opinion related expression assigned the opinion contribution degree in the opinion related expression storage device;
Opinion-related expression end determination means for checking whether or not there is an unprocessed opinion-related expression stored in the opinion-related expression storage device;
A sentence acquisition means for acquiring a full sentence from the sentence storage device when the opinion related expression end determination means determines an end;
For the sentence acquired by the sentence acquisition means, the opinion likelihood of the sentence is calculated from the opinion contribution of the opinion related expression appearing in the sentence stored in the opinion related expression storage device, and the sentence 2. The opinion sentence search device according to claim 1, further comprising: an opinion likeness index calculation unit that stores the opinion likeness in the opinion storage index-added sentence storage device. In a method for outputting an opinion sentence related to a search word from a plurality of sentences,
A search term input step in which the search term input means inputs a search term;
A search result sentence acquisition unit that acquires a sentence in which the search word input in the search word input step appears from a sentence storage device in which a plurality of sentences are stored;
Opinion-likeness index assigning means assigns an opinion-likeness index to every sentence of the sentence storage device, and stores all sentences with the opinion-likeness index in the sentence storage device with the opinion-likeness index A uniqueness index assigning step;
Opinion-likeness index required for the opinion-sentence acquisition means to determine the degree of opinion sentence stored in the sentence storage device with the opinion-likeness index from the search result sentence acquired by the search result sentence acquisition step. Based on the above, an opinion sentence acquisition step for acquiring an opinion sentence with an index of opinion likeness,
An opinion sentence search method, wherein the opinion sentence output means includes an opinion sentence output step of outputting the opinion sentence acquired by the opinion sentence acquisition step based on the opinion likelihood index. The opinion-likeness index assigning step includes
Opinion related expression acquisition means for acquiring one opinion related expression from an opinion related expression storage device in which expressions related to opinions are stored;
Opinion-likeness index acquisition means acquires a sentence including the one opinion-related expression acquired by the opinion-related expression acquisition step and an opinion-like index assigned to the sentence from the sentence storage device with an opinion-likeness index. An opinion-likeness acquisition step,
The degree to which the opinion-related expression in the sentence affects the opinion contribution degree calculation means from the opinion-like index acquired in the opinion-like index acquisition step affects the determination of the sentence as an opinion. An opinion contribution degree calculating step of calculating an opinion contribution degree to be indicated, and storing the opinion related expression assigned the opinion contribution degree in the opinion related expression storage device;
Opinion-related expression end determination means is stored in the opinion-related expression storage device, and the opinion-related expression end determination step for checking whether or not there is an unprocessed opinion-related expression in the opinion contribution degree granting process;
If the sentence acquisition unit is determined to be ended by the opinion related expression end determination step, a sentence acquisition step of acquiring a full sentence from the sentence storage device;
Opinion-likeness index calculation means, for the sentence acquired by the sentence acquisition step, from the opinion contribution degree of the opinion-related expression appearing in one sentence stored in the opinion-related expression storage device, The opinion sentence search method according to claim 3, further comprising: calculating an opinion likeness and storing the sentence and the opinion likeness in a sentence storage device with an opinion likeness index.   An opinion sentence search program characterized in that it is a program for causing a computer to execute the opinion sentence search method according to claim 3.   6. A recording medium recording an opinion sentence search program, wherein the program according to claim 5 is recorded on a recording medium readable by the computer.

Images