JP5729633B2 - Terminal device, expression output method, and program - Google Patents

Description

  The present invention relates to a terminal device that extracts an evaluation expression from a sentence.

  In the first conventional technique, in order to summarize the reputation information included in the review article by the consumer, the reputation information is classified according to the characteristics that the reputation information is focused on. In this method, for example, in the case of the topic “digital camera”, classification can be made for each feature such as “image quality” and “size”, and positive or negative opinions can be known (see Non-Patent Document 1). ).

  In the second conventional technique, there is a reputation information classification device that can classify reputation information for each topic even if the evaluation expressions are the same, if related topics are different (see Patent Document 1). Even if the evaluation expression is the same, the apparatus performs clustering using the peripheral text of the evaluation expression to classify the evaluation expression for each topic if the related topics are different.

JP 2010-198278 A (first page, FIG. 1 etc.)

Minqing Hu and Bing Lie: Mining Opinion Features in Customer Reviews, AAAI 2004 (2004)

  However, in the conventional technology, it is not possible to output conflicting evaluation expressions from a large amount of documents such as the Web. Note that the conflicting evaluation expressions are evaluation expressions that describe the same viewpoint and are sets of evaluation expressions that have conflicting contents. For example, the positive expression “xylitol is a sweetener that does not cause caries” is a negative expression for the topic “xylitol”, whereas it can be caries with xylitol gum. Since the opposite content is stated for the same viewpoint, it is a conflicting evaluation expression. On the other hand, positive expressions such as “Xylitol is a safe sweetener” or “Xylitol is cheap” are the expressions with the opposite evaluation polarity described on the same topic as xylitol. These are not conflicting evaluation expressions, because the viewpoints that are concerned are safety and price and not about the efficacy of preventing tooth decay.

That is, in the first technique, it is necessary to create a dictionary lists whether the advance what features exist. In addition, it is difficult to handle evaluation information that is not clear as to what characteristics are to be evaluated, such as “a sweetener that does not cause caries”. Therefore, in the first technique, it is not possible to extract a conflicting evaluation expression with a high degree of freedom without any preparation.
In other words, for example, as an expression opposite to negative expression 1 “May be caries due to xylitol gum”, only positive expression related to the same topic “Xylitol is a sweetener that does not cause caries”. Rather, it also extracts positive expressions on different topics, such as "Xylitol is a sweetener that does not cause caries", "Xylitol gum keeps teeth strong and healthy", and "Xylitol is cheap." Can do.

  The second technique does not consider extracting conflicting evaluation expressions and cannot extract conflicting evaluation expressions.

In view of the above problems, the present invention has the following solutions.
The terminal device according to the first aspect of the present invention is a terminal device that outputs both a positive expression and a negative expression related to one topic from sentences stored in one or more server devices. And evaluation phrase dictionary that can store one or more evaluation phrase information having evaluation polarity that is information indicating whether the evaluation phrase is positive or negative, and stored in one or more server devices Two or more evaluation expressions that are a set of one or more words including evaluation words, using the evaluation words stored in the evaluation word dictionary from sentences related to the topic. Two or more evaluation expressions are clustered from one or more words / phrases included in each of the two or more evaluation expressions extracted by the evaluation expression extraction unit and the evaluation expression extraction unit to be extracted. Evaluation expression An evaluation expression clustering unit that obtains a loop and two or more evaluation expressions included in one evaluation expression group of one or more evaluation expression groups, using an evaluation word and evaluation polarity included in the evaluation word dictionary , A confrontation evaluation expression acquisition unit that obtains a positive evaluation expression related to one topic and a negative evaluation expression related to one topic, and a positive evaluation expression and negative regarding one topic acquired by the confrontation evaluation expression acquisition unit It is a terminal device provided with the opposite evaluation expression output part which outputs both typical evaluation expressions.

  With this configuration, it is possible to output conflicting evaluation expressions from a large amount of documents such as the Web.

  In addition, the terminal device of the second aspect of the present invention is different from the first aspect in that one or more words / phrases included in the evaluation expression, the original form of each phrase, the part of speech of each phrase, the evaluation polarity of each phrase and the phrase Kind information indicating the type of each phrase indicating whether it is a starting phrase of the evaluation expression, a phrase in the middle of the evaluation expression, or a phrase unrelated to the evaluation expression, and a sequence of one or more words included in the evaluation expression A learning data storage unit capable of storing two or more pieces of learning data for each evaluation expression, and the evaluation expression extraction unit constitutes sentences stored in one or more server devices. Morphological analysis of each sentence to be obtained, morpheme analysis means for acquiring one or more words, and from the processing result of the morpheme analysis means, one or more words acquired by the morpheme analysis means, the original form of each word, the part of speech of each word, And included in the evaluation expression A feature acquisition unit that acquires information related to an arrangement of one or more words, searches the evaluation word dictionary using one or more words as a key, and acquires evaluation polarities of the one or more words for each sentence; Using information acquired by the acquisition means as a feature, machine learning is performed using two or more learning data in the learning data storage unit, and information on one or more columns of type information indicating the types of words constituting each sentence is acquired. 1 corresponding to a type information sequence in which type information indicating a start phrase of an evaluation expression or a phrase in the middle of an evaluation expression is continuous among machine learning means and one or more types of information columns acquired by the machine learning means. It is a terminal device comprising an evaluation expression acquisition means for acquiring an evaluation expression that is a set of the above continuous phrases.

  With this configuration, it is possible to appropriately extract evaluation expressions from a large amount of documents such as the Web.

  Further, in the terminal device according to the third aspect of the present invention, in contrast to the first aspect, the evaluation expression clustering unit includes one or more words included in each of the two or more evaluation expressions extracted by the evaluation expression extraction unit. An attribute expression vector for each evaluation expression is constructed from the phrase appearance information acquisition means for acquiring appearance information, which is information relating to the appearance, for each evaluation expression, and the appearance information for each evaluation expression acquired by the phrase appearance information acquisition means. Using the attribute vector construction means, the similarity calculation means for calculating the similarity between the evaluation expressions using the attribute vector for each evaluation expression formed by the attribute vector construction means, and the similarity calculated by the similarity calculation means An evaluation expression group acquiring unit that clusters two or more evaluation expressions and acquires two or more evaluation expression groups including the one or more evaluation expressions.

  With this configuration, evaluation expressions can be appropriately clustered.

  In the terminal device of the fourth aspect of the invention, in the third aspect of the invention, the attribute vector has an appearance frequency in the evaluation expression for one or more words (w) included in the evaluation expression. tf (w) Using the frequency cf (w) included in all evaluation expressions extracted from sentences stored in one or more server devices, the dimension corresponding to each word is “tf (w) / The terminal device is a vector having a value of “cf (w)”.

  With this configuration, evaluation expressions can be appropriately clustered.

  In the terminal device according to the fifth aspect of the present invention, in contrast to the third or fourth aspect, the evaluation expression clustering unit further includes a high-frequency evaluation phrase storage unit that can store one or more evaluation words that are frequently used. The phrase appearance information acquisition means includes a weight of appearance information corresponding to the evaluation phrase stored in the high-frequency evaluation phrase storage means smaller than a weight of appearance information corresponding to another phrase that is not the evaluation phrase. As described above, the terminal device acquires, for each evaluation expression, appearance information of one or more words included in each of the two or more evaluation expressions extracted by the evaluation expression extraction unit.

  With this configuration, evaluation expressions can be clustered more appropriately.

  Further, in the terminal device according to the sixth aspect of the invention, in contrast to the first aspect, the confrontation evaluation expression acquisition unit is an evaluation polarity string that is an evaluation polarity string composed of evaluation expressions, and is included in the evaluation expression The evaluation polarity sequence that is a sequence of evaluation polarities corresponding to one or more words, one or more words included in the evaluation expression, the original form of each word, the part of speech of each word, and the evaluation expression are positive. One of the evaluation expression learning data storage means that can store two or more evaluation expression learning data that associates evaluation expression polarity that is information indicating whether or not, and one or more evaluation expression groups Two or more evaluation expressions included in the evaluation expression group are divided into words, one or more words are acquired, and the phrase information acquisition means for acquiring the original form and part of speech of each of one or more words, and the phrase information acquisition One or more words acquired by the means as a key The evaluation word dictionary is searched, the evaluation polarity corresponding to each word is acquired, the evaluation polarity string acquisition means for acquiring the evaluation polarity string that is one or more evaluation polarity strings for each evaluation expression, and the phrase information acquisition Two or more evaluation expressions stored in the evaluation expression learning data storage means, each of the one or more words acquired by the means, the original form of each word, the part of speech of each word, and the evaluation polarity string acquired by the evaluation polarity string acquisition means Applies to learning data using machine learning, and the evaluation expression polarity acquisition means for acquiring the evaluation expression polarity of each evaluation expression for each evaluation expression, and the evaluation expression polarity acquired by the evaluation expression polarity acquisition means is positive. The terminal device includes an evaluation expression and a conflict evaluation expression acquisition unit that acquires an evaluation expression that is negative in the evaluation expression polarity acquired by the evaluation expression polarity acquisition unit.

  With this configuration, it is possible to appropriately extract conflicting evaluation expressions from a large amount of documents such as the Web.

  Further, the terminal device according to the seventh aspect of the present invention further comprises a receiving unit for receiving a topic from the user with respect to any one of the first to sixth aspects, and the evaluation expression extracting unit is stored in one or more server devices. An evaluation that is a set of one or more words including evaluation words using sentences that are included in the evaluation word stored in the evaluation word dictionary from sentences related to the topic received by the reception unit. A terminal device that extracts two or more expressions.

  With this configuration, it is possible to output a conflicting evaluation expression that is an expression related to an input topic from a large amount of documents such as the Web.

  The terminal device according to the present invention can output conflicting evaluation expressions from a large amount of documents such as the Web.

Block diagram of the information system in the first embodiment Block diagram of the terminal device Flow chart for explaining the operation of the terminal device Flow chart for explaining the operation of the evaluation expression extraction process A flowchart for explaining the operation of the evaluation expression clustering process A flowchart for explaining the operation of the appearance information calculation process Flow chart for explaining the operation of the grouping process A flowchart for explaining the operation of the conflict evaluation expression acquisition process The figure which shows the evaluation word information management table Figure showing the same learning data management table The figure which shows the same high frequency evaluation word / phrase management table Figure showing the evaluation expression learning data management table The figure which shows the example of the information stored in the text storage part The figure explaining the process which groups the evaluation expression group The figure which shows the output example of the terminal device Block diagram of terminal apparatus according to Embodiment 2 Overview of the computer system in the above embodiment Block diagram of the computer system

Hereinafter, embodiments of a terminal device and the like will be described with reference to the drawings. In addition, since the component which attached | subjected the same code | symbol in embodiment performs the same operation | movement, description may be abbreviate | omitted again.
(Embodiment 1)

  In this embodiment, an information system that extracts and outputs both positive expressions and negative expressions from a large amount of documents such as the Web will be described.

  FIG. 1 is a block diagram of an information system 1 in the present embodiment. FIG. 2 is a block diagram of the terminal device 12 constituting the information system 1.

  The information system 1 includes one or more server devices 11 and one or more terminal devices 12. Each server device 11 and each terminal device 12 can usually communicate with each other via a network such as the Internet. The server device 11 is a device that stores text, for example, a Web server device. The terminal device 12 is a device that outputs both a positive expression and a negative expression related to one topic from sentences stored in one or more server devices 11.

  The server device 11 includes a text storage unit 111 and a text transmission unit 112.

  The terminal device 12 includes a text storage unit 120, an evaluation phrase dictionary 121, a learning data storage unit 122, a reception unit 123, an evaluation expression extraction unit 124, an evaluation expression clustering unit 125, an opponent evaluation expression acquisition unit 126, and an opponent evaluation expression output unit. 127. In the terminal device 12, the evaluation expression extraction unit 124 extracts the evaluation expression corresponding to the topic received from the reception unit 123. At this time, the evaluation expression extraction unit 124 uses information of the text storage unit 120, the evaluation phrase dictionary 121, and the learning data storage unit 122. In addition, the evaluation expression clustering unit 125 clusters the evaluation expressions extracted by the evaluation expression extraction unit 124. Further, from the clustered evaluation expressions, the conflict evaluation expression acquisition unit 126 acquires a conflict evaluation expression, and the conflict evaluation expression output unit 127 outputs the conflict evaluation expression.

  The evaluation expression extraction unit 124 includes a morpheme analysis unit 1241, a feature acquisition unit 1242, a machine learning unit 1243, and an evaluation expression acquisition unit 1244.

  The evaluation expression clustering unit 125 includes high-frequency evaluation phrase storage means 1251, phrase appearance information acquisition means 1252, attribute vector construction means 1253, similarity calculation means 1254, and evaluation expression group acquisition means 1255.

  The confrontation evaluation expression acquisition unit 126 includes evaluation expression learning data storage means 1261, phrase information acquisition means 1262, evaluation polarity string acquisition means 1263, evaluation expression polarity acquisition means 1264, and confrontation evaluation expression acquisition means 1265.

  The sentence storage unit 111 of the server device 11 stores one or more sentences. The text is usually a file, but the data format such as a database is not limited. Moreover, the structure of a sentence does not ask | require HTML, XML, plain text, etc. The text is, for example, a web page.

  The text transmission unit 112 transmits the text in the text storage unit 111 to the terminal device 12. Normally, the text transmission unit 112 transmits text in response to a request from the terminal device 12.

  The text storage unit 120 stores one or more sentences. The sentence is a sentence received from the server device 11. The text is received by, for example, a receiving means (not shown). The text storage unit 120 may temporarily store a sentence.

  The evaluation phrase dictionary 121 of the terminal device 12 can store one or more pieces of evaluation phrase information. The evaluation phrase information is information having an evaluation phrase and an evaluation polarity. The evaluation phrase is a phrase related to evaluation. The evaluation phrase information includes, for example, “excellent”, “like”, “good”, “dislike”, “bad”, “good”, “tasteful”, and the like. The evaluation polarity is information indicating whether the evaluation phrase is positive or negative. The evaluation polarity may be, for example, information indicating either affirmative / negative or information of any of three or more types. The evaluation polarity may be, for example, information from “+5” to “−5”. Here, “+5” is very positive, “−5” is very negative, etc. A phrase is usually a word, but it may be a morpheme or phrase and is widely understood. That is, the phrase may be a set of two or more words. Furthermore, when the evaluation polarity is information indicating either positive or negative, for example, positive is “1” and negative “−1”.

  The learning data storage unit 122 can store two or more learning data. The learning data includes, for example, one or more words / phrases included in the evaluation expression, the original form of each phrase, the part of speech of each phrase (“part of speech of phrase” is equivalent to “quality of word”), and evaluation of each phrase. This is data (information) having information on polarity, type information, and an arrangement of one or more words included in the evaluation expression. The type information is a word (for example, “B”) of the evaluation expression or a word in the middle of the evaluation expression (for example, “I”), or a phrase (for example, irrelevant to the evaluation expression). This is information indicating the type of each word indicating whether it is “O”. The learning data is data for each evaluation expression. Note that the learning data is not limited to the information described above as long as it is information related to the evaluation expression. It is preferable that the learning data is information relating to words constituting the evaluation expression.

  The reception unit 123 receives a topic. A topic is usually a phrase. However, the topic may be a sentence including a phrase. When the reception unit 123 receives a sentence including a phrase, the phrase may be extracted from the sentence, and the extracted phrase may be passed to the evaluation expression extraction unit 124. Here, reception means reception of information input from an input device such as a keyboard, mouse, touch panel, reception of information transmitted via a wired or wireless communication line, recording on an optical disk, magnetic disk, semiconductor memory, or the like. It is a concept including reception of information read from a medium.

  The topic input means may be anything such as a numeric keypad, keyboard, mouse or menu screen. The receiving unit 123 can be realized by a device driver for input means such as a numeric keypad or a keyboard, control software for a menu screen, and the like.

  The evaluation expression extraction unit 124 is a sentence that constitutes a sentence stored in one or more server apparatuses 11, and extracts two or more evaluation expressions from a sentence related to a topic. An evaluation expression is a set of one or more words including evaluation words. The evaluation expression extraction unit 124 extracts two or more evaluation expressions from the sentence related to the topic using the evaluation words stored in the evaluation word dictionary 121. Here, the topic is usually a topic received by the receiving unit 123. Further, the evaluation expression extraction unit 124 may include a receiving unit that receives text from one or more server devices 11. As a technique for realizing the evaluation expression extraction unit 124, the technique described in “Tetsuji Nakagawa, Takuya Kawada, Kentaro Inui and Sadao Kurohashi: Extracting Subjective and Objective Evaluative Expressions from the Web, ISUC-2008 (2008)” is used. be able to. A sentence related to a topic is a sentence including a topic, a topic and a synonym, a synonym, a broader word, a narrower word, a sentence that can be acquired from the topic, or the like.

  The morpheme analyzing means 1241 constituting the evaluation expression extracting unit 124 performs morphological analysis on each sentence constituting the sentence stored in the one or more server apparatuses 11, and acquires one or more words / phrases. The morpheme analyzing unit 1241 may perform morphological analysis on each sentence constituting the sentence, and obtain part of speech of one or more words, original forms of each word, and the like. Note that the morphological analysis unit 1241 may perform morphological analysis and acquire the original form of each acquired phrase from a word dictionary (not shown). Moreover, since morphological analysis is a well-known technique, detailed description is abbreviate | omitted. Moreover, since the technique for acquiring the original form of the phrase is also a known technique, detailed description thereof is omitted.

  The feature acquisition unit 1242 uses the processing result of the morpheme analysis unit 1241, for example, one or more each phrase acquired by the morpheme analysis unit 1241, the original form of each phrase, the part of speech of each phrase, and one or more included in the evaluation expression The information on the arrangement of the phrases is acquired, the evaluation phrase dictionary 121 is searched using one or more words as a key, and the evaluation polarity of each of the one or more phrases is acquired for each sentence. For example, when the sentence is “xylitol is a sweetener that does not cause caries”, the information regarding the arrangement of one or more words included in the evaluation expression is “xylitol | "Sweetizer |". "|" Is a phrase delimiter. The evaluation polarity of one or more words acquired for each sentence is, for example, “0 | 0 | -1 | 0 | 0 | 0 | 0 | 0”. Here, “0” is a word that is neither positive nor negative, “1” is a positive word, and “−1” is a negative word. The feature acquisition unit 1242 acquires the feature used by the machine learning unit 1243. The feature is information relating to the phrase, and is not limited to the information described above. Moreover, it is not essential to include all the information described above in the feature.

  The machine learning unit 1243 uses the information acquired by the feature acquisition unit 1242 as a feature, performs machine learning using two or more learning data in the learning data storage unit 122, and stores type information indicating the types of words constituting each sentence. Get information about one or more columns. Here, the information regarding one or more columns of the type information is, for example, “O | O | B | I | I | I | I | I”. Here, “O” is a word irrelevant to the evaluation expression, “B” is a start word of the evaluation expression, and “I” is a word in the middle of the evaluation expression. The machine learning may use any algorithm such as SVM (support vector machine) or decision tree. Since machine learning is a known technique, a detailed description thereof will be omitted.

  The evaluation expression acquisition means 1244 is a type information string in which type information indicating that it is a start phrase of an evaluation expression or a phrase in the middle of an evaluation expression among one or more types of information strings acquired by the machine learning means 1243. An evaluation expression that is a set of one or more corresponding consecutive phrases is acquired. The evaluation expression acquisition unit 1244 acquires, for example, a type information string “BIIIII” that is a column of “B” or “I” in “O | O | B | I | I | I | I | I”, An evaluation expression “It is a sweetener that does not cause cavity” that is a set of words corresponding to “BIIIII” is acquired.

  The evaluation expression clustering unit 125 acquires two or more evaluation expression groups including one or more evaluation expressions. That is, the evaluation expression clustering unit 125 classifies the two or more evaluation expressions into two or more groups (evaluation expression groups) based on the commonality of the phrases included in each evaluation expression. Specifically, the evaluation expression clustering unit 125 clusters two or more evaluation expressions from one or more words / phrases included in each of the two or more evaluation expressions extracted by the evaluation expression extraction unit 124, and performs one or more evaluations. Obtain two or more evaluation expression groups including expressions. The evaluation expression clustering unit 125 acquires two or more evaluation expression groups including one or more evaluation expressions. Specifically, the evaluation expression clustering unit 125 calculates the similarity between two or more evaluation expressions from one or more words included in each of the two or more evaluation expressions extracted by the evaluation expression extraction unit 124, From the similarity, it is preferable to cluster two or more evaluation expressions and obtain two or more evaluation expression groups including one or more evaluation expressions. Here, “acquisition of evaluation expression group” is to classify evaluation expressions. “Acquiring an evaluation expression group” may be to assign a group identifier to each evaluation expression, or to accumulate each evaluation expression in a storage area divided by group. That is, it is only necessary to classify evaluation expressions into groups. As a technology for realizing the evaluation expression clustering unit 125, “Toshihiro Kamisu: Clustering method in the data mining field (1)-Let's use clustering!-, Journal of Artificial Intelligence, Vol.18, No.1, pp .59-65 (2003). "Can also be used. Moreover, the evaluation expression clustering unit 125 classifies the evaluation expressions into two or more groups based on the proximity of the attribute vectors using the attribute vectors corresponding to the evaluation expressions.

  The high-frequency evaluation phrase storage unit 1251 that constitutes the evaluation expression clustering unit 125 can store one or more evaluation phrases that are frequently used. One or more evaluation phrases that are frequently used are, for example, “excellent”, “dislike”, and “like” that are often used.

  The phrase appearance information acquisition unit 1252 acquires, for each evaluation expression, the appearance information of one or more words included in each of the two or more evaluation expressions extracted by the evaluation expression extraction unit 124. The appearance information is information related to the appearance of a word. The appearance information includes, for example, the number of appearances, the appearance ratio (number of appearances / number of all words / phrases), “tf (w) / cf (w)” described later, and the like.

  Moreover, the phrase appearance information acquisition unit 1252 makes the weight of the appearance information corresponding to the evaluation word stored in the high-frequency evaluation word storage unit 1251 smaller than the weight of the appearance information corresponding to another word that is not the evaluation word. As such, the appearance information of one or more words included in each of the two or more evaluation expressions extracted by the evaluation expression extraction unit 124 may be acquired for each evaluation expression. Here, the other words are words that are not stored in the high-frequency evaluation word storage unit 1251. “To make the weight of appearance information small” means, for example, that the appearance information (value) is multiplied by “0.8” or the appearance information (value) is “0.2”. This can be realized by performing a process such as subtracting.

  The attribute vector construction unit 1253 constructs an evaluation expression attribute vector for each evaluation expression from the appearance information for each evaluation expression acquired by the phrase appearance information acquisition unit 1252. The attribute vector is, for example, a set of (word / phrase, appearance information). The attribute vector is, for example, the appearance frequency tf (w) of each word (w) in the evaluation expression and one or more servers for one or more words (w) included in the evaluation expression. Using the frequency cf (w) of each word (w) included in all evaluation expressions extracted from sentences stored in the apparatus 11, the dimension corresponding to each word is “tf (w) / cf ( w) ".

  The similarity calculating unit 1254 calculates the similarity between the evaluation expressions using the attribute vector for each evaluation expression configured by the attribute vector configuring unit 1253. Here, the similarity is, for example, a value calculated using the Euclidean distance between vectors. Also, the similarity may be, for example, a degree of similarity calculated using the number of matching vector elements (a value calculated by an increasing function using the number of matching vector elements as a parameter) or the like. There can be various methods for calculating the similarity between attribute vectors.

  The evaluation expression group acquisition unit 1255 clusters two or more evaluation expressions using the similarity calculated by the similarity calculation unit 1254, and acquires two or more evaluation expression groups including one or more evaluation expressions. Here, “using similarity” means, for example, that evaluation expressions whose similarity is within a threshold value are grouped into the same group. Also, the evaluation expression group acquisition unit 1255 makes a pair of evaluation expressions having the closest attribute vector corresponding to each evaluation expression, and recursively creates a pair of evaluation expression groups with the pair as one evaluation expression group. When the above condition (the number of recursive processes or the similarity is within a threshold) is satisfied, the process is terminated, and two or more evaluation expressions are classified into groups.

  The confrontation evaluation expression acquisition unit 126 uses an evaluation word and an evaluation polarity included in the evaluation word dictionary 121 from two or more evaluation expressions included in one evaluation expression group of one or more evaluation expression groups. Get a positive rating expression for a topic and a negative rating expression for a topic.

  The evaluation expression learning data storage unit 1261 constituting the conflict evaluation expression acquisition unit 126 can store two or more evaluation expression learning data. The evaluation expression learning data is, for example, data that associates an evaluation polarity string, each word constituting the evaluation expression, the original form of each word, the part of speech of each word, and the evaluation expression polarity. The evaluation expression learning data storage unit 1261 can store two or more evaluation expression learning data. The evaluation expression learning data is preferably data having, for example, an evaluation polarity string, each word constituting the evaluation expression, the original form of each word, the part of speech of each word, and the evaluation expression polarity. . The evaluation polarity column is a column of evaluation polarity composed of evaluation expressions, and is an array of evaluation polarities corresponding to one or more words / phrases included in the evaluation expression. The evaluation expression polarity is information indicating whether the evaluation expression is positive or negative. The evaluation expression learning data may include information other than the evaluation polarity string, each word, the original form of each word, the part of speech of each word, and the evaluation expression polarity, and it is essential that all of the above information is included. is not.

  The phrase information acquisition unit 1262 acquires phrase information corresponding to two or more evaluation expressions included in any evaluation expression group of one or more evaluation expression groups. The phrase information is information related to one or more words constituting the evaluation expression, and is, for example, one or more words constituting the evaluation expression, the original form and the part of speech of each of the one or more words. The phrase information acquisition unit 1262 divides each of the two or more evaluation expressions included in any one of the two or more evaluation expression groups into words, acquires one or more words, and Get the original form and part of speech of each of one or more words. Also, the phrase information acquisition unit 1262 divides the evaluation expression into phrases using, for example, a morphological analysis technique, and acquires the original form and part of speech of the phrase. Needless to say, the phrase information acquisition unit 1262 may use each phrase acquired by the feature acquisition unit 1242, the original form of each phrase, and the part of speech of each phrase. In addition, the phrase information acquisition unit 1262 may search a word dictionary (not shown) using the phrase that is the result of the morphological analysis as a key, and acquire the original form.

  The evaluation polarity string acquisition unit 1263 searches the evaluation phrase dictionary 121 using one or more words acquired by the phrase information acquisition unit 1262 as a key, acquires one or more evaluation polarities corresponding to each phrase, An evaluation polarity column that is a column of evaluation polarities is acquired for each evaluation expression. If the word is not an evaluation word, the evaluation polarity string acquisition unit 1263 cannot acquire the evaluation polarity corresponding to the word.

  The evaluation expression polarity acquisition unit 1264 applies the evaluation polarity sequence acquired by the evaluation polarity sequence acquisition unit 1263 to two or more evaluation expression learning data stored in the evaluation expression learning data storage unit 1261 using machine learning. The evaluation expression polarity of the evaluation polarity column is acquired for each evaluation expression.

  The confrontation evaluation expression acquisition unit 1265 includes one or more evaluation expressions in which the evaluation expression polarity acquired by the evaluation expression polarity acquisition unit 1264 is positive, and one or more in which the evaluation expression polarity acquired by the evaluation expression polarity acquisition unit is negative. Get an evaluation expression of. When the conflict evaluation expression acquisition unit 1265 selects one positive or negative evaluation expression from a number of positive or negative evaluation expressions, it may be arbitrarily selected, or the length of the evaluation expression, You may select based on attribute values, such as the number of words.

  The conflict evaluation expression output unit 127 outputs both a positive evaluation expression and a negative evaluation expression related to one topic acquired by the conflict evaluation expression acquisition unit 126. It goes without saying that outputting the evaluation expression includes outputting a sentence including the evaluation expression and outputting a file including the evaluation expression. Here, output refers to display on a display, projection using a projector, printing on a printer, sound output, transmission to an external device, storage in a recording medium, output to another processing device or other program, etc. It is a concept that includes delivery of processing results.

  The sentence storage unit 111, the text storage unit 120, the evaluation phrase dictionary 121, the learning data storage unit 122, the high-frequency evaluation phrase storage unit 1251, and the evaluation expression learning data storage unit 1261 are preferably non-volatile recording media. It is also possible to realize the recording medium with the same characteristics. In addition, the process in which evaluation phrase information etc. are memorize | stored in the evaluation phrase dictionary 121 grade | etc., Is not ask | required. For example, evaluation phrase information or the like may be stored in the evaluation phrase dictionary 121 or the like via a recording medium, and evaluation phrase information or the like transmitted via a communication line or the like is stored in the evaluation phrase dictionary 121 or the like. Alternatively, evaluation phrase information or the like input via the input device may be stored in the evaluation phrase dictionary 121 or the like.

  The text transmitting unit 112 is usually realized by a wireless or wired communication unit, but may be realized by a broadcasting unit.

  Evaluation expression extraction unit 124, evaluation expression clustering unit 125, confrontation evaluation expression acquisition unit 126, morpheme analysis unit 1241, feature acquisition unit 1242, machine learning unit 1243, evaluation expression acquisition unit 1244, phrase appearance information acquisition unit 1252, attribute vector configuration The means 1253, the similarity calculation means 1254, the evaluation expression group acquisition means 1255, the phrase information acquisition means 1262, the evaluation polarity string acquisition means 1263, the evaluation expression polarity acquisition means 1264, and the conflict evaluation expression acquisition means 1265 are usually MPU, memory, etc. Can be realized from. The processing procedure of the evaluation expression extraction unit 124 or the like is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The conflict evaluation expression output unit 127 may or may not include an output device such as a display or a speaker. The conflict evaluation expression output unit 127 can be realized by driver software of an output device or driver software and an output device of an output device.

  Next, the operation of the information system 1 will be described. First, the operation of the server device 11 will be described. The text transmission unit 112 of the server device 11 transmits the text in the text storage unit 111 to the terminal device 12 via the network 13 in response to a request from the terminal device 12. The terminal device 12 may request the server device 11 to transmit a sentence when the receiving unit 123 receives a topic, or requests the server apparatus 11 to transmit a sentence in accordance with a sentence transmission request from the user. You may do it. The sentence transmitting unit 112 may send only sentences related to the topic, may send only sentences including sentences related to the topic, or may send all stored sentences. .

  Next, operation | movement of the terminal device 12 is demonstrated using the flowchart of FIG. In the flowchart of FIG. 3, it is assumed that one or more sentences are already stored in the text storage unit 120.

  (Step S301) The receiving unit 123 determines whether a topic has been received. If a topic is accepted, the process goes to step S302, and if no topic is accepted, the process returns to step S301.

  (Step S302) The evaluation expression extraction unit 124 extracts all sentences related to the topic received in step S301 from the text storage unit 120.

  (Step S303) The evaluation expression extraction unit 124 extracts an evaluation expression from each sentence read in step S302. Such evaluation expression extraction processing will be described with reference to the flowchart of FIG.

  (Step S304) The evaluation expression clustering unit 125 classifies the evaluation expressions extracted in step S303 into groups. Such evaluation expression clustering processing will be described with reference to the flowchart of FIG.

  (Step S305) The confrontation evaluation expression acquisition unit 126 acquires confrontation evaluation expressions from two or more evaluation expressions included in one of the groups classified in Step S304. The conflict evaluation expression acquisition process will be described with reference to the flowchart of FIG.

  (Step S306) The conflict evaluation expression output unit 127 outputs the conflicting evaluation expression acquired in step S305, and returns to step S301.

  In the flowchart of FIG. 3, the process ends when the power is turned off or the process is terminated.

  Next, the evaluation expression extraction process in step S303 will be described with reference to the flowchart of FIG.

  (Step S401) The evaluation expression extraction unit 124 substitutes 1 for a counter i.

  (Step S402) The evaluation expression extraction unit 124 determines whether or not the i-th sentence exists. If the i-th sentence exists, it will go to step S403, and if it does not exist, it will return to a high-order process (step S304).

  (Step S403) The morpheme analyzing unit 1241 acquires the i-th sentence.

  (Step S404) The morpheme analyzing unit 1241 performs morphological analysis on the sentence acquired in Step S403, and acquires one or more words (or morpheme).

  (Step S405) The feature acquisition unit 1242 substitutes 1 for a counter j.

  (Step S406) The feature acquisition unit 1242 determines whether or not the j-th word is present in the words acquired in step S404. If the jth word / phrase exists, the process goes to step S407, and if the jth word / phrase does not exist, the process goes to step S410.

  (Step S407) The feature acquisition unit 1242 includes one or more words / phrases acquired by the morpheme analysis unit 1241, the original form of each word, the part of speech of each word, and the evaluation expression using the processing result of the morpheme analysis in Step S404. Get information about the sequence of one or more words. Then, the feature acquisition unit 1242 adds the acquired information to a buffer (not shown).

  (Step S408) The feature acquisition unit 1242 searches the evaluation word dictionary 121 using the j-th word as a key, and acquires the evaluation polarity of the word. Then, the feature acquisition unit 1242 additionally writes the acquired evaluation polarity in a buffer (not shown) in association with the information added in step S407.

  (Step S409) The feature acquisition unit 1242 increments the counter j by 1, and returns to step S406.

  (Step S410) The machine learning means 1243 obtains one or more words / phrases stored in the buffer, the original form of each word / phrase, the part of speech of each word / phrase, information on the arrangement of one or more words / phrases included in the evaluation expression, and the evaluation polarity. Machine learning is performed using two or more pieces of learning data stored in the learning data storage unit 122 as features. Then, the machine learning unit 1243 obtains one or more columns of type information corresponding to the column of phrases that are the result of the machine learning.

  (Step S411) The evaluation expression acquisition unit 1244 is a type in which type information indicating that it is a start phrase of the evaluation expression or a word in the middle of the evaluation expression among the one or more types of information column acquired in Step S410. Try to get the information column.

  (Step S412) The evaluation expression acquisition unit 1244 determines whether or not the type information string has been acquired in step S411. If the type information string can be acquired, the process proceeds to step S413. If the type information string cannot be acquired, the process jumps to step S414.

  (Step S413) The evaluation expression acquisition unit 1244 acquires an evaluation expression that is one or more words corresponding to each type information included in the type information string.

  (Step S414) The evaluation expression extraction unit 124 increments the counter i by 1, and returns to step S402.

  Next, the evaluation expression clustering process in step S304 will be described using the flowchart of FIG.

  (Step S501) The evaluation expression clustering unit 125 substitutes 1 for a counter i.

  (Step S502) The evaluation expression clustering unit 125 determines whether or not the i-th evaluation expression exists among the evaluation expressions extracted by the evaluation expression extraction process. If the i-th evaluation expression exists, the process goes to step S503, and if the i-th evaluation expression does not exist, the process goes to step S510.

  (Step S503) The phrase appearance information acquisition unit 1252 divides the i-th evaluation expression into phrases. The phrase appearance information acquisition unit 1252 may perform morphological analysis on the i th evaluation expression and divide it into words, or one or more already divided phrases are stored in a storage medium (not shown) in association with the evaluation expression. One or more words may be read out.

  (Step S504) The phrase appearance information acquisition unit 1252 substitutes 1 for the counter j.

  (Step S505) The phrase appearance information acquisition unit 1252 determines whether or not the j-th phrase exists. If the jth word / phrase exists, the process goes to step S506, and if the jth word / phrase does not exist, the process goes to step S508.

  (Step S506) The phrase appearance information acquisition unit 1252 performs a process of calculating appearance information. Such appearance information calculation processing will be described with reference to the flowchart of FIG. Note that the calculated appearance information is at least temporarily stored in the buffer in association with the j-th word.

  (Step S507) The phrase appearance information acquisition unit 1252 increments the counter j by 1, and returns to Step S505.

  (Step S508) The attribute vector construction unit 1253 constructs an attribute vector of the i-th evaluation expression from the appearance information for each word acquired in Step S506. That is, for example, the attribute vector construction unit 1253 constructs an attribute vector (appearance information of phrase 1, appearance information of phrase 2,..., Appearance information of phrase n).

  (Step S509) The evaluation expression clustering unit 125 increments the counter i by 1, and returns to step S502.

  (Step S510) The similarity calculation unit 1254 and the evaluation expression group acquisition unit 1255 classify two or more evaluation expressions into groups using the attribute vector, and return to the higher-level processing (step S305). Such grouping processing will be described with reference to the flowchart of FIG.

  Next, the appearance information calculation process in step S506 will be described with reference to the flowchart of FIG.

  (Step S601) The phrase appearance information acquisition unit 1252 acquires the phrase (w).

  (Step S602) The phrase appearance information acquisition unit 1252 acquires the appearance frequency tf (w) of the phrase (w) in the evaluation expression of interest.

  (Step S603) The phrase appearance information acquisition unit 1252 acquires the appearance frequency cf (w) of the phrase (w) in all the evaluation expressions.

  (Step S604) The phrase appearance information acquisition unit 1252 calculates “appearance information = tf (w) / cf (w)”.

  (Step S605) The phrase appearance information acquisition unit 1252 searches the frequently evaluated phrase storage unit 1251 using the phrase (w) as a key, and determines whether or not the phrase (w) is stored in the frequently evaluated phrase storage unit 1251. to decide. If it is stored in the high-frequency evaluation phrase storage means 1251, the process goes to step S606, and if it is not stored, the process jumps to step S607. Note that the phrase stored in the frequently evaluated phrase storage unit 1251 is a frequently evaluated phrase.

  (Step S606) The phrase appearance information acquisition unit 1252 performs a process of reducing the appearance information calculated in Step S604.

  (Step S607) The phrase appearance information acquisition unit 1252 accumulates the appearance information calculated in Step S604 or Step S606 in the buffer in association with the phrase (w), and returns to the higher-level process (Step S507).

  In the flowchart of FIG. 6, the process of reducing the appearance information in step S606 may be performed on tf (w) calculated in step S602, for example.

  Next, the grouping process of step S510 will be described using the flowchart of FIG.

  (Step S701) The similarity calculation unit 1254 substitutes 1 for a counter i.

  (Step S702) The similarity calculation unit 1254 determines whether or not the i-th evaluation expression exists. If the i-th evaluation expression exists, the process goes to step S703, and if not, the process goes to step S711.

  (Step S703) The similarity calculation unit 1254 acquires an attribute vector corresponding to the i-th evaluation expression.

  (Step S704) The similarity calculation means 1254 substitutes 1 for the counter j.

  (Step S705) The similarity calculation unit 1254 determines whether or not there is another j-th evaluation expression (an evaluation expression that is not an i-th evaluation expression but an evaluation expression that has been paired in the past). If it exists, go to step S706, otherwise go to step S709.

  (Step S706) The similarity calculation means 1254 acquires an attribute vector corresponding to the jth evaluation expression.

  (Step S707) The similarity calculation unit 1254 calculates the similarity between the attribute vector corresponding to the i th evaluation expression and the attribute vector corresponding to the j th evaluation expression. Note that the similarity is, for example, a value (for example, a distance between 1 / vectors) calculated by a decreasing function using a distance between vectors as a parameter.

  (Step S708) The similarity calculation unit 1254 increments the counter j by 1, and returns to step S705.

  (Step S709) The similarity calculation unit 1254 sets the evaluation expression corresponding to the attribute vector having the highest similarity to the attribute vector corresponding to the i-th evaluation expression as a pair of the i-th evaluation expression. “Pairing” means that two evaluation expressions can be associated with each other by some information, such as linking two evaluation expressions.

  (Step S710) The similarity calculation unit 1254 increments the counter i by 1, and returns to step S702.

  (Step S711) The evaluation expression clustering unit 125 performs group determination processing using the paired evaluation expressions as one evaluation expression. “Making a pair of evaluation expressions as one evaluation expression” means, for example, generating one attribute vector from each attribute vector of the paired evaluation expressions. “Generate one attribute vector from two or more attribute vectors” means that an average value of each element of two or more attribute vectors may be calculated for each element, or two or more attribute vectors are connected. It may be a single vector.

  (Step S712) The evaluation expression clustering unit 125 determines whether or not all evaluation expressions are in one group. If it becomes one group, it will go to step S713, and if it will not become one group, it will return to step S711.

  (Step S713) The evaluation expression group acquiring unit 1255 sets two or more evaluation expressions within a predetermined similarity range as one group. For example, the evaluation expression group acquisition unit 1255 gives the same group identifier to two or more evaluation expressions within a predetermined similarity range. Then, the process returns to the upper process (step S305).

  In the flowchart of FIG. 7, for example, the process may be terminated when a group having a similarity with a similarity equal to or less than a threshold is formed.

  Next, the conflict evaluation expression acquisition process of step S305 will be described using the flowchart of FIG.

  (Step S801) The confrontation evaluation expression acquisition unit 126 acquires two or more evaluation expressions of one group.

  (Step S802) The conflict evaluation expression acquisition unit 126 substitutes 1 for a counter i.

  (Step S803) The phrase information acquisition unit 1262 determines whether or not the i-th evaluation expression exists. If it exists, go to step S804, and if not, go to step S813.

  (Step S804) The phrase information acquisition unit 1262 acquires the i-th evaluation expression, and divides the evaluation expression into one or more words. Note that the phrase information acquisition unit 1262 may read one or more divided phrases corresponding to the i-th evaluation expression.

  (Step S805) The evaluation polarity string acquisition unit 1263 substitutes 1 for the counter j.

  (Step S806) The evaluation polarity string acquisition unit 1263 determines whether or not the j-th word / phrase exists in the i-th evaluation expression. If it exists, go to step S807, and if not, go to step S810.

  (Step S807) The evaluation polarity string acquisition unit 1263 searches the evaluation word dictionary 121 using the jth word as a key, and acquires the evaluation polarity corresponding to the jth word. Then, the evaluation polarity string acquisition unit 1263 temporarily stores the acquired evaluation polarity in the buffer in association with the j-th word / phrase. In addition, accumulation here is additional writing.

  (Step S808) The phrase information acquisition unit 1262 acquires the original form and part of speech of the jth phrase.

  (Step S809) The counter j is incremented by 1, and the process returns to step S806.

  (Step S810) The evaluation expression polarity acquisition unit 1264 stores, in the evaluation expression learning data storage unit 1261, the evaluation polarity string, each phrase, the original form of each phrase, the part of speech of each phrase, and the like accumulated in the buffer in step S807. Applying to two or more evaluation expression learning data using machine learning, the evaluation expression polarity of the evaluation expression is acquired.

  (Step S811) The evaluation expression polarity acquisition means 1264 associates the i-th evaluation expression with the evaluation expression polarity acquired in step S809, and stores them in a recording medium (not shown).

  (Step S812) The conflict evaluation expression acquiring unit 126 increments the counter i by 1, and the process returns to Step S803.

  (Step S813) The confrontation evaluation expression acquisition unit 1265 acquires one or more (usually one) evaluation expressions whose evaluation expression polarity acquired by the evaluation expression polarity acquisition unit 1264 is positive and the evaluation expression polarity acquisition unit. One or more (usually one) evaluation expressions having negative evaluation expression polarity are obtained. Then, the process returns to the upper process (step S306).

  Hereinafter, a specific operation of the information system 1 in the present embodiment will be described.

  Now, it is assumed that one or more server apparatuses 11 store various Web pages such as blogs and news articles.

  Further, it is assumed that the evaluation phrase dictionary 121 of the terminal device 12 stores the evaluation phrase information management table shown in FIG. The evaluation phrase information management table has one or more evaluation phrase information. The evaluation phrase information includes “ID”, “evaluation phrase”, and “evaluation polarity”. An evaluation phrase is a term indicating evaluation of an object. The evaluation polarity “1” indicates a positive evaluation, and “−1” indicates a negative evaluation.

  Further, it is assumed that the learning data storage unit 122 stores a learning data management table shown in FIG. The learning data management table stores one or more records having “ID”, “evaluation expression”, “phrase”, “original form”, “part of speech”, “evaluation polarity”, and “type information”. Note that the order of words in the record is information regarding the arrangement of one or more words included in the evaluation expression.

  Further, the high-frequency evaluation phrase storage unit 1251 holds a high-frequency evaluation word management table shown in FIG. The high frequency evaluation phrase management table stores one or more records having “ID” and “high frequency evaluation phrase”.

  Further, the evaluation expression learning data storage unit 1261 holds an evaluation expression learning data management table shown in FIG. The evaluation expression learning data management table stores one or more records having “ID”, “evaluation expression polarity”, “evaluation polarity string”, “phrase”, “original form”, “part of speech”, and “evaluation expression”.

  In this situation, it is assumed that the user inputs the topic “xylitol” to the terminal device 12. Then, the reception unit 123 of the terminal device 12 receives “xylitol”.

  Next, the reception unit 123 prompts one or more server apparatuses 11 to transmit a sentence including “xylitol”.

  Next, the sentence transmission unit 112 of each of the one or more server apparatuses 11 receives a transmission request, and reads a sentence including “xylitol” from the sentence storage unit 111. Then, the sentence transmitting unit 112 transmits a sentence including “xylitol” to the terminal device 12.

  Next, a receiving unit (not shown) of the terminal device 12 (which may be a receiving unit included in the receiving unit 123) receives sentences including “xylitol” from one or more server apparatuses 11 and accumulates them in the text storage unit 120. Assume that a set of sentences shown in FIG. 13 is stored in the text storage unit 120, for example.

  This process is similar to the process of a so-called Web search engine, and is a known technique, and thus detailed description thereof is omitted.

  Next, the evaluation expression extraction unit 124 reads all sentences including the topic “xylitol” from the text storage unit 120. Then, the evaluation expression extraction unit 124 extracts evaluation expressions from all the read sentences as follows.

  That is, first, the morphological analysis unit 1241 of the evaluation expression extraction unit 124 performs morphological analysis on each sentence and acquires one or more words. For example, in the case of a sentence “xylitol is a sweetener that does not cause caries”, the morpheme analyzing means 1241 uses, for example, the phrase “xylitol | is | caries | , Morpheme).

  Next, the feature acquisition unit 1242 acquires information on one or more words acquired by the morpheme analysis unit 1241, the original form of each word, the part of speech of each word, and the arrangement of one or more words included in the evaluation expression. That is, for example, for the sentence “xylitol is a sweetener that does not cause caries”, the feature acquisition means 1242 uses each word “xylitol | is | cavity || is awakened | not | sweetener | is”, each word The original form of “xylitol | ha | mushi | | wake | not | sweetener | is”, the part of speech “noun | particle” | noun | particle | verb | auxiliary | noun | auxiliary verb ”of each phrase is acquired. Note that the information related to the arrangement of one or more words is “xylitol | is | cavity | | awake | not | sweetener | is”.

  Next, the feature acquisition unit 1242 searches the evaluation word dictionary 121 (see FIG. 9) using each word as a key, and acquires the evaluation polarity of the word. For example, for the sentence “xylitol is a sweetener that does not cause caries”, the feature acquisition means 1242 determines the evaluation polarity corresponding to “xylitol | is | Column "0 | 0 | -1 | 0 | 0 | 0 | 0 | 0" is acquired.

  Through the processing described above, the feature acquisition unit 1242 performs, for each sentence, one or more words / phrases, the original form of each word / phrase, the part of speech of each word / phrase, information on the arrangement of one or more words / phrases included in the evaluation expression, Evaluation polarity was obtained.

  Next, the machine learning unit 1243 uses one or more words / phrases stored in the buffer, the original form of each word / phrase, the part of speech of each word / phrase, information on the arrangement of one or more words / phrases included in the evaluation expression, and the evaluation polarity as features. Then, machine learning is performed using two or more pieces of learning data stored in the learning data storage unit 122 (FIG. 10). Then, the machine learning unit 1243 obtains information on one or more columns of type information indicating the types of words constituting each sentence. For example, it is assumed that the machine learning means 1243 has acquired “O | O | B | I | I | I | I | I” for the sentence “xylitol is a sweetener that does not cause caries”. .

  Next, the evaluation expression acquisition unit 1244 continues the type information indicating that it is the start phrase “B” of the evaluation expression or the phrase “I” in the middle of the evaluation expression among the one or more columns of the acquired type information. Type information string “B | I | I | I | I | I” is acquired. Then, the evaluation expression acquisition unit 1244 acquires an evaluation expression “It is a sweetener that does not cause caries” that is one or more words corresponding to the type information string “B | I | I | I | I | I”. Such processing is performed for all sentences, and a large number of evaluation expressions are acquired. In addition, from the sentence "There is a possibility of becoming a cavity with xylitol gum", the evaluation expression "It may become a cavity" is obtained, and from the sentence "Xylitol gum keeps teeth strong and healthy" Assume that the evaluation expression “Keep teeth healthy and healthy” has been acquired.

  Next, the evaluation expression clustering unit 125 classifies a large number of extracted evaluation expressions into groups as follows.

  First, the phrase appearance information acquisition unit 1252 divides each evaluation expression into phrases. For example, for the evaluation expression “It is a sweetener that does not cause tooth decay”, the phrase appearance information acquisition unit 1252 divides the cavity into “words |

  Next, the phrase appearance information acquisition unit 1252 calculates appearance information for each phrase. That is, first, the phrase appearance information acquisition unit 1252 acquires the phrase (w) (for example, “cavities”). The phrase appearance information acquisition unit 1252 acquires the appearance frequency tf (w) “1” of the phrase (w) “cavities” in the target evaluation expression “a sweetener that does not cause cavity”. Next, the phrase appearance information acquisition unit 1252 acquires the appearance frequency cf (w) (for example, “15”) of the phrase (w) “cavities” in all the evaluation expressions. Next, the phrase appearance information acquisition unit 1252 calculates “appearance information = tf (w) / cf (w) = 1/15”. Next, the phrase appearance information acquisition unit 1252 searches the high-frequency evaluation phrase storage unit 1251 using the phrase (w) “cavity” as a key, and the phrase (w) is not stored in the high-frequency evaluation phrase storage unit 1251. Judge. Then, the phrase appearance information acquisition unit 1252 stores the calculated appearance information “1/15” in the buffer in association with the phrase (w) “cavities”. The above processing is performed for each word.

  Next, the attribute vector construction unit 1253 constructs an attribute vector of each evaluation expression from the acquired appearance information for each word. For example, for the evaluation expression “It is a sweetener that does not cause caries”, the attribute vector construction means 1253 has (1 /,..., 1/15,..., 1/20,. ). Each element of the attribute vector is usually appearance information of all the terms. For example, the number of elements in the attribute vector matches the number of types of words.

  Next, the similarity calculation unit 1254 and the evaluation expression group acquisition unit 1255 classify two or more evaluation expressions into groups using attribute vectors of each evaluation expression as follows.

  That is, the similarity calculation unit 1254 determines the attribute vector having the highest similarity to the attribute vector of each evaluation expression, and groups the two evaluation expressions corresponding to the two attribute vectors. At that time, the similarity calculation unit 1254 also stores the similarity in the buffer in association with the two evaluation expressions. Note that, as described above, the similarity calculation unit 1254 calculates the similarity between two attribute vectors by a decreasing function (for example, inversely proportional) using the distance between the two attribute vectors as a parameter.

  Next, the evaluation expression clustering unit 125 regards the paired evaluation expressions as one evaluation expression, and performs group determination processing recursively. That is, for example, when evaluation expressions 1 to 8 exist, as shown in FIG. 14, the two evaluation expression groups having the largest similarity among the evaluation expression groups that are paired (one type of group) are selected. Group up into groups. Then, for example, the evaluation expression group acquisition unit 1255 sets the threshold “0.5” as a threshold for dividing the group. That is, in FIG. 14, the evaluation expressions are grouped by the line having the threshold value “0.5”. Then, the evaluation expression clustering unit 125 determines the eight evaluation expressions as the evaluation expression 1, the evaluation expression 2, the evaluation expression 3, the first group having the evaluation expression 4, the evaluation expression 5, and the second group having the evaluation expression 6. The third group having the expression 7 and the evaluation expression 8 is classified into three groups. That is, for example, the evaluation expression group acquisition unit 1255 gives the first group identifier “1” to the evaluation expression 1, the evaluation expression 2, the evaluation expression 3, and the evaluation expression 4, and the evaluation expression 5 and the evaluation expression 6 The second group identifier “2” is assigned, and the third group identifier “3” is assigned to the evaluation expressions 7 and 8. For example, it is assumed that the evaluation expression 1 is “a sweetener that does not cause a cavity” and the evaluation expression 2 is “possibly a cavity”. That is, for example, the first group is a group of evaluation expressions related to “cavities”.

  Next, the confrontation evaluation expression acquisition unit 126 acquires confrontation evaluation expressions as follows. That is, the conflict evaluation expression acquisition unit 126 acquires, for example, four evaluation expressions to which the first group identifier “1” is assigned. Here, the conflict evaluation expression acquisition unit 126 may arbitrarily select the first group, or may select the first group according to a user instruction.

  Next, the phrase information acquisition unit 1262 divides each evaluation expression into one or more phrases. The phrase information acquisition unit 1262 obtains, for example, “a toothpaste | is not awakened | not a sweetener | is” from “a sweetener that does not cause a cavity”. In addition, the phrase information acquisition unit 1262 obtains, for example, “possible to become a cavity” from “possibly a cavity” |

  Next, the evaluation polarity string acquisition unit 1263 searches the evaluation word dictionary 121 (see FIG. 9) using each word of each evaluation expression as a key, and acquires the evaluation polarity corresponding to each word. For example, it is assumed that the evaluation polarity string acquisition unit 1263 has acquired “−1 | 0 | 0 | 0 | 0 | 0” for “it is | raised | not | sweetener | is”. Also, the evaluation polarity string acquisition unit 1263 acquires “−1 | 0 | 0 | 0 | 0 | 0 | 0” in response to “possibility | is | Suppose that

  Next, the phrase information acquisition unit 1262 acquires the original form and part of speech of each phrase. For example, the evaluation polarity string acquisition means 1263, with respect to “cavities | is | raised | not | sweetener | is”, the original form of each phrase “cavities | is | raised | not | sweetener | is”, Get the part of speech "noun | particle | verb | particle | noun | particle". Further, for example, in contrast to “the tooth || | | | possibility | is | | |”, the phrase information acquisition means 1262 has the original form “cavity | | | | | possibility | is | The part of speech “noun | particle” | verb | noun | particle | verb is acquired for each phrase.

  Next, the evaluation expression polarity acquisition unit 1264 stores two or more evaluations stored in the evaluation expression learning data storage unit 1261 such as the evaluation polarity string, each word, the original form of each word, the part of speech of each word, and the like accumulated in the buffer. It applies to expression learning data (refer FIG. 12) using machine learning, and the evaluation expression polarity of each evaluation expression is acquired. The evaluation expression polarity acquisition unit 1264 acquires, for example, the evaluation expression polarity “+” for the evaluation expression “a sweetener that does not cause tooth decay” and the evaluation expression polarity “−” for “possibly a cavity”. Here, the evaluation expression polarity “+” is positive, and “−” is negative.

  Next, the evaluation expression polarity acquisition unit 1264 associates each evaluation expression with the acquired evaluation expression polarity and stores them in a recording medium (not shown). That is, for example, the evaluation expression polarity acquisition unit 1264 accumulates “It is a sweetener that does not cause a cavity, +” and “It may become a cavity, −” in a recording medium.

  Next, the confrontation evaluation expression acquisition unit 1265 has one evaluation expression (for example, “a sweetener that does not cause caries”) having a positive evaluation expression polarity and the evaluation expression polarity acquired by the evaluation expression polarity acquisition unit. One evaluation expression that is negative (for example, “it may become a cavity”) is acquired. Here, it is assumed that the conflict evaluation expression acquisition unit 1265 has acquired the sentences “xylitol is a sweetener that does not cause caries” and “it may become a caries due to xylitol gum” including the evaluation expressions.

  Next, the conflict evaluation expression output unit 127 outputs the acquired conflict evaluation expression. Here, it is assumed that the conflict evaluation expression output unit 127 outputs the sentences including the evaluation expression “xylitol is a sweetener that does not cause caries” and “it may become caries due to xylitol gum”. . An example of such output is shown in FIG.

As described above, according to the present embodiment, it is possible to appropriately output conflicting evaluation expressions from a large amount of documents such as the Web. Therefore, according to the present embodiment, for example, it is possible to realize a search engine that outputs a conflict opinion regarding an input topic.
Further, according to the present embodiment, it is possible to extract conflicting evaluation expressions without designating features or viewpoints in advance.

  In the present embodiment, it is preferable to use machine learning to extract evaluation expressions, but other algorithms such as pattern matching using evaluation words “good” and “bad” may be used. The same applies to other embodiments.

  Further, in the present embodiment, it is preferable to use the similarity of attribute vectors for clustering evaluation expressions, but the nouns (evaluation targets) constituting the evaluation expressions are grouped into the same group, etc. Other algorithms may be used. The same applies to other embodiments.

  In the present embodiment, high-frequency evaluation words / phrases are used for clustering evaluation expressions, but high-frequency evaluation words / phrases may not be used. The same applies to other embodiments.

  In the present embodiment, it is preferable to use machine learning to acquire the characteristics (evaluation polarity) of the evaluation expression, but the evaluation expression is obtained from the result of multiplying the evaluation polarities of the evaluation words constituting the evaluation expression. Other algorithms may be used, such as obtaining the characteristics of Multiplying the evaluation polarity indicates, for example, that negation of negation is affirmative and affirmation denial is negation. The same applies to other embodiments.

Furthermore, the processing in the present embodiment may be realized by software. Then, this software may be distributed by software download or the like. Further, this software may be recorded and distributed on a recording medium such as a CD-ROM. This also applies to other embodiments in this specification. The software that realizes the terminal device in the present embodiment is the following program. In other words, this program has one or more sentences, an evaluation phrase that is a word relating to evaluation, and an evaluation polarity that is information indicating whether the evaluation word is positive or negative on the recording medium. The above evaluation word / phrase information is stored, the computer is a sentence constituting the sentence stored in the recording medium, and the evaluation word / phrase stored in the recording medium is used from the sentence related to the topic The evaluation expression extraction unit that extracts two or more evaluation expressions that are a set of one or more words including the evaluation word and the one or more included in each of the two or more evaluation expressions extracted by the evaluation expression extraction unit An evaluation expression that calculates a similarity between two or more evaluation expressions from the words, and clusters the two or more evaluation expressions from the similarity to obtain one or more evaluation expression groups including one or more evaluation expressions Comment From the expression clustering unit and two or more evaluation expressions included in any one of the one or more evaluation expression groups, an evaluation phrase and an evaluation polarity included in the recording medium are used. Opposite evaluation expression acquisition unit that acquires both positive evaluation expression and negative evaluation expression related to a topic, and positive evaluation expression and negative evaluation expression related to the one topic acquired by the opposing evaluation expression acquisition unit Is a program for functioning as a conflict evaluation expression output unit that outputs both.
(Embodiment 2)

  In the present embodiment, a mode in which the processing of the system of the first embodiment is performed by one terminal device will be described. In the present embodiment, a terminal device that outputs both positive expressions and negative expressions related to one topic will be described. The terminal device 2 in the present embodiment is different from the terminal device 12 constituting the information system 1 of the first embodiment in that it includes a text storage unit 111.

  FIG. 16 is a block diagram of terminal device 2 in the present embodiment.

  The terminal device 2 includes a text storage unit 111, an evaluation phrase dictionary 121, a learning data storage unit 122, a reception unit 123, an evaluation expression extraction unit 124, an evaluation expression clustering unit 125, an opponent evaluation expression acquisition unit 126, and an opponent evaluation expression output unit 127. Is provided. Each component constituting the terminal device 2 has been described in the first embodiment.

  Note that the evaluation expression extraction unit 124 also performs processing for acquiring a sentence related to the topic received by the reception unit 123 from the sentence in the sentence storage unit 111.

  Next, the operation of the terminal device 2 will be described. Since the operation of the terminal device 2 is substantially the same as the operation of the terminal device 12 described in the first embodiment, the description thereof is omitted.

  As described above, according to this embodiment, it is possible to appropriately output conflicting evaluation expressions from a large amount of documents.

  FIG. 17 shows the external appearance of a computer that executes the program described in this specification to realize the terminal device or the like of the above-described embodiment. The above-described embodiments can be realized by computer hardware and a computer program executed thereon. FIG. 17 is an overview diagram of the computer system 340, and FIG. 18 is a diagram illustrating an internal configuration of the computer system 340.

  In FIG. 17, the computer system 340 includes a computer 341 including an FD drive 3411 and a CD-ROM drive 3412, a keyboard 342, a mouse 343, and a monitor 344.

  In FIG. 18, in addition to the FD drive 3411 and the CD-ROM drive 3412, the computer 341 stores an MPU 3413, a bus 3414 connected to the CD-ROM drive 3412 and the FD drive 3411, and a program such as a bootup program. A RAM 3416 for temporarily storing application program instructions and providing a temporary storage space; and a hard disk 3417 for storing application programs, system programs, and data. Although not shown here, the computer 341 may further include a network card that provides connection to the LAN.

  A program that causes the computer system 340 to execute the functions of the terminal device or the like of the above-described embodiment is stored in the CD-ROM 3501 or FD 3502, inserted into the CD-ROM drive 3412 or FD drive 3411, and further stored in the hard disk 3417. May be forwarded. Alternatively, the program may be transmitted to the computer 341 via a network (not shown) and stored in the hard disk 3417. The program is loaded into the RAM 3416 at the time of execution. The program may be loaded directly from the CD-ROM 3501, the FD 3502, or the network.

  The program does not necessarily include an operating system (OS), a third-party program, or the like that causes the computer 341 to execute the functions of the terminal device according to the above-described embodiment. The program only needs to include an instruction portion that calls an appropriate function (module) in a controlled manner and obtains a desired result. How the computer system 340 operates is well known and will not be described in detail.

  In the above program, in a step of transmitting information, a step of receiving information, etc., processing performed by hardware, for example, processing performed by a modem or an interface card (processing performed only by hardware) is performed. Not included.

  Further, the computer that executes the program may be singular or plural. That is, centralized processing may be performed, or distributed processing may be performed.

  Further, in each of the above embodiments, it goes without saying that two or more communication units existing in one apparatus may be physically realized by one medium.

  In each of the above embodiments, each process (each function) may be realized by centralized processing by a single device (system), or by distributed processing by a plurality of devices. May be.

  The present invention is not limited to the above-described embodiments, and various modifications are possible, and it goes without saying that these are also included in the scope of the present invention.

  As described above, the terminal device according to the present invention has an effect of being able to output conflicting evaluation expressions from a large amount of documents such as the Web, and implements a search engine that outputs conflicting opinions regarding the input topic. It is useful as a device.

DESCRIPTION OF SYMBOLS 1 Information system 2, 12 Terminal device 11 Server apparatus 111 Text storage part 112 Text transmission part 120 Text storage part 121 Evaluation phrase dictionary 122 Learning data storage part 123 Reception part 124 Evaluation expression extraction part 125 Evaluation expression clustering part 126 Acquisition of conflict evaluation expression Unit 127 Conflict evaluation expression output unit 1241 Morphological analysis unit 1242 Feature acquisition unit 1243 Machine learning unit 1244 Evaluation expression acquisition unit 1251 High-frequency evaluation phrase storage unit 1252 Phrase appearance information acquisition unit 1253 Attribute vector configuration unit 1254 Similarity calculation unit 1255 Evaluation expression Group acquisition means 1261 Evaluation expression learning data storage means 1262 Phrase information acquisition means 1263 Evaluation polarity string acquisition means 1264 Evaluation expression polarity acquisition means 1265 Opposition evaluation expression acquisition means

Claims (8)

A terminal device that outputs both positive and negative expressions related to one topic from sentences stored in one or more server devices,
An evaluation phrase dictionary that can store one or more evaluation phrase information having evaluation phrases that are evaluation-related phrases and evaluation polarity that is information indicating whether the evaluation phrase is positive or negative;
A sentence constituting a sentence stored in the one or more server devices, and using an evaluation phrase stored in the evaluation phrase dictionary from a sentence related to the topic, the one or more including the evaluation phrase An evaluation expression extraction unit that extracts two or more evaluation expressions that are sets of phrases;
The similarity between two or more evaluation expressions is calculated from one or more words included in each of the two or more evaluation expressions extracted by the evaluation expression extracting unit, and the two or more evaluation expressions are calculated from the similarity. An evaluation expression clustering unit that performs clustering and obtains one or more evaluation expression groups including two or more evaluation expressions;
Using one or more evaluation expressions included in one evaluation expression group among the one or more evaluation expression groups, an evaluation word and evaluation polarity included in the evaluation word dictionary and an affirmative regarding the one topic A conflict evaluation expression acquisition unit that acquires a negative evaluation expression and a negative evaluation expression related to the one topic;
A terminal device comprising: a conflict evaluation expression output unit that outputs both a positive evaluation expression and a negative evaluation expression related to the one topic acquired by the conflict evaluation expression acquisition unit ;
One or more words included in the evaluation expression, the original form of each word, the part of speech of each word, the evaluation polarity of each word, and whether the word is a starting word or a word in the middle of the evaluation expression Learning data having type information indicating the type of each of the phrases indicating whether the phrase is irrelevant to the evaluation expression, and information relating to an arrangement of one or more words included in the evaluation expression, and learning data for each evaluation expression Two or more learning data storage units that can be stored,
The evaluation expression extraction unit includes:
A morpheme analysis unit that obtains one or more words by morphologically analyzing each sentence constituting a sentence stored in the one or more server devices;
From the processing result of the morpheme analysis unit, obtain information on one or more words acquired by the morpheme analysis unit, the original form of each of the words, the part of speech of each of the words, and the arrangement of one or more words included in the evaluation expression And a feature acquisition unit that searches the evaluation word dictionary using one or more words as a key, and acquires an evaluation polarity of each of the one or more words for each sentence;
Information on one or more columns of type information indicating the types of words constituting each sentence by machine learning using information acquired by the feature acquisition means as features and using two or more learning data in the learning data storage unit Machine learning means for acquiring
Among the one or more types of information column acquired by the machine learning means, one or more continuous types corresponding to the type information sequence in which the type information indicating that the evaluation expression is a starting word or phrase in the middle of the evaluation expression is continuous. A terminal device comprising evaluation expression acquisition means for acquiring an evaluation expression that is a set of phrases . A terminal device that outputs both positive and negative expressions related to one topic from sentences stored in one or more server devices,
An evaluation phrase dictionary that can store one or more evaluation phrase information having evaluation phrases that are evaluation-related phrases and evaluation polarity that is information indicating whether the evaluation phrase is positive or negative;
A sentence constituting a sentence stored in the one or more server devices, and using an evaluation phrase stored in the evaluation phrase dictionary from a sentence related to the topic, the one or more including the evaluation phrase An evaluation expression extraction unit that extracts two or more evaluation expressions that are sets of phrases;
The similarity between two or more evaluation expressions is calculated from one or more words included in each of the two or more evaluation expressions extracted by the evaluation expression extracting unit, and the two or more evaluation expressions are calculated from the similarity. An evaluation expression clustering unit that performs clustering and obtains one or more evaluation expression groups including two or more evaluation expressions;
Using one or more evaluation expressions included in one evaluation expression group among the one or more evaluation expression groups, an evaluation word and evaluation polarity included in the evaluation word dictionary and an affirmative regarding the one topic A conflict evaluation expression acquisition unit that acquires a negative evaluation expression and a negative evaluation expression related to the one topic;
A terminal device comprising: a conflict evaluation expression output unit that outputs both a positive evaluation expression and a negative evaluation expression related to the one topic acquired by the conflict evaluation expression acquisition unit;
The conflict evaluation expression acquisition unit
An evaluation polarity string that is a string of evaluation polarities composed of evaluation expressions, an evaluation polarity string that is an array of evaluation polarities corresponding to one or more words included in the evaluation expression, and 1 included in the evaluation expression An evaluation expression that associates each of the above phrases, the original form of each of the phrases, the part of speech of each of the phrases, and an evaluation expression polarity that is information indicating whether the evaluation expression is positive or negative Evaluation expression learning data storage means capable of storing two or more learning data;
Two or more evaluation expressions included in any one of the one or more evaluation expression groups are divided into words, one or more words are obtained, and the original form of each of the one or more words and Phrase information acquisition means for acquiring a part of speech;
Using the one or more words acquired by the word information acquisition means as a key, the evaluation word dictionary is searched, an evaluation polarity corresponding to each word is acquired, and an evaluation polarity column that is a column of the one or more evaluation polarities is obtained. Evaluation polarity string acquisition means for acquiring for each evaluation expression;
One or more each phrase acquired by the phrase information acquisition unit, the original form of each phrase, the part of speech of each phrase, and the evaluation polarity string acquired by the evaluation polarity string acquisition unit are stored in the evaluation expression learning data storage unit. An evaluation expression polarity acquisition unit that applies machine learning to two or more evaluation expression learning data, and acquires the evaluation expression polarity of each evaluation expression for each evaluation expression;
Opposite evaluation expression acquisition means for acquiring an evaluation expression in which the evaluation expression polarity acquired by the evaluation expression polarity acquisition means is positive and an evaluation expression in which the evaluation expression polarity acquired by the evaluation expression polarity acquisition means is negative; A terminal device comprising: A terminal device that outputs both positive and negative expressions related to a topic,
A sentence storage unit capable of storing one or more sentences;
An evaluation phrase dictionary that can store one or more evaluation phrase information having evaluation phrases that are evaluation-related phrases and evaluation polarity that is information indicating whether the evaluation phrase is positive or negative;
A sentence constituting a sentence stored in the sentence storage unit, and an evaluation phrase stored in the evaluation phrase dictionary from a sentence related to the topic, and one or more words including the evaluation phrase An evaluation expression extraction unit that extracts two or more evaluation expressions as a set; and
The similarity between two or more evaluation expressions is calculated from one or more words included in each of the two or more evaluation expressions extracted by the evaluation expression extracting unit, and the two or more evaluation expressions are calculated from the similarity. An evaluation expression evaluation expression clustering unit that performs clustering and obtains one or more evaluation expression groups including one or more evaluation expressions;
Using one or more evaluation expressions included in one evaluation expression group among the one or more evaluation expression groups, an evaluation word and evaluation polarity included in the evaluation word dictionary and an affirmative regarding the one topic A conflict evaluation expression acquisition unit that acquires a negative evaluation expression and a negative evaluation expression related to the one topic;
A terminal device comprising: a conflict evaluation expression output unit that outputs both a positive evaluation expression and a negative evaluation expression related to the one topic acquired by the conflict evaluation expression acquisition unit ;
One or more words included in the evaluation expression, the original form of each word, the part of speech of each word, the evaluation polarity of each word, and whether the word is a starting word or a word in the middle of the evaluation expression Learning data having type information indicating the type of each of the phrases indicating whether the phrase is irrelevant to the evaluation expression, and information relating to an arrangement of one or more words included in the evaluation expression, and learning data for each evaluation expression Two or more learning data storage units that can be stored,
The evaluation expression extraction unit includes:
Morphological analysis of each sentence constituting the one or more sentences and obtaining one or more words;
From the processing result of the morpheme analysis unit, obtain information on one or more words acquired by the morpheme analysis unit, the original form of each of the words, the part of speech of each of the words, and the arrangement of one or more words included in the evaluation expression And a feature acquisition unit that searches the evaluation word dictionary using one or more words as a key, and acquires an evaluation polarity of each of the one or more words for each sentence;
Information on one or more columns of type information indicating the types of words constituting each sentence by machine learning using information acquired by the feature acquisition means as features and using two or more learning data in the learning data storage unit Machine learning means for acquiring
Among the one or more types of information column acquired by the machine learning means, one or more continuous types corresponding to the type information sequence in which the type information indicating that the evaluation expression is a starting word or phrase in the middle of the evaluation expression is continuous. A terminal device comprising evaluation expression acquisition means for acquiring an evaluation expression that is a set of phrases . A terminal device that outputs both positive and negative expressions related to a topic,
A sentence storage unit capable of storing one or more sentences;
An evaluation phrase dictionary that can store one or more evaluation phrase information having evaluation phrases that are evaluation-related phrases and evaluation polarity that is information indicating whether the evaluation phrase is positive or negative;
A sentence constituting a sentence stored in the sentence storage unit, and an evaluation phrase stored in the evaluation phrase dictionary from a sentence related to the topic, and one or more words including the evaluation phrase An evaluation expression extraction unit that extracts two or more evaluation expressions as a set; and
The similarity between two or more evaluation expressions is calculated from one or more words included in each of the two or more evaluation expressions extracted by the evaluation expression extracting unit, and the two or more evaluation expressions are calculated from the similarity. An evaluation expression evaluation expression clustering unit that performs clustering and obtains one or more evaluation expression groups including one or more evaluation expressions;
Using one or more evaluation expressions included in one evaluation expression group among the one or more evaluation expression groups, an evaluation word and evaluation polarity included in the evaluation word dictionary and an affirmative regarding the one topic A conflict evaluation expression acquisition unit that acquires a negative evaluation expression and a negative evaluation expression related to the one topic;
A terminal device comprising: a conflict evaluation expression output unit that outputs both a positive evaluation expression and a negative evaluation expression related to the one topic acquired by the conflict evaluation expression acquisition unit;
The conflict evaluation expression acquisition unit
An evaluation polarity string that is a string of evaluation polarities composed of evaluation expressions, an evaluation polarity string that is an array of evaluation polarities corresponding to one or more words included in the evaluation expression, and 1 included in the evaluation expression An evaluation expression that associates each of the above phrases, the original form of each of the phrases, the part of speech of each of the phrases, and an evaluation expression polarity that is information indicating whether the evaluation expression is positive or negative Evaluation expression learning data storage means capable of storing two or more learning data;
Two or more evaluation expressions included in any one of the one or more evaluation expression groups are divided into words, one or more words are obtained, and the original form of each of the one or more words and Phrase information acquisition means for acquiring a part of speech;
Using the one or more words acquired by the word information acquisition means as a key, the evaluation word dictionary is searched, an evaluation polarity corresponding to each word is acquired, and an evaluation polarity column that is a column of the one or more evaluation polarities is obtained. Evaluation polarity string acquisition means for acquiring for each evaluation expression;
One or more each phrase acquired by the phrase information acquisition unit, the original form of each phrase, the part of speech of each phrase, and the evaluation polarity string acquired by the evaluation polarity string acquisition unit are stored in the evaluation expression learning data storage unit. An evaluation expression polarity acquisition unit that applies machine learning to two or more evaluation expression learning data, and acquires the evaluation expression polarity of each evaluation expression for each evaluation expression;
Opposite evaluation expression acquisition means for acquiring an evaluation expression in which the evaluation expression polarity acquired by the evaluation expression polarity acquisition means is positive and an evaluation expression in which the evaluation expression polarity acquired by the evaluation expression polarity acquisition means is negative; A terminal device comprising: On the recording medium,
One or more sentences,
Storing one or more evaluation word / phrase information having an evaluation word / phrase relating to evaluation and an evaluation polarity which is information indicating whether the evaluation word / phrase is positive or negative;
An expression output method realized by an evaluation expression extraction unit, an evaluation expression evaluation expression clustering unit, an opponent evaluation expression acquisition unit, and an opponent evaluation expression output unit,
The evaluation expression extraction unit is a sentence that constitutes a sentence stored in the recording medium, and includes the evaluation phrase using an evaluation phrase stored in the recording medium from a sentence related to the topic An evaluation expression extraction step of extracting two or more evaluation expressions that are a set of one or more words;
The evaluation expression evaluation expression clustering unit clusters the two or more evaluation expressions from one or more words / phrases included in each of the two or more evaluation expressions extracted in the evaluation expression extraction step. An evaluation expression evaluation expression clustering step for obtaining one or more evaluation expression groups including:
The conflict evaluation expression acquisition unit uses an evaluation word and an evaluation polarity included in the recording medium from two or more evaluation expressions included in any evaluation expression group of the one or more evaluation expression groups, An opposing evaluation expression acquisition step of acquiring both a positive evaluation expression and a negative evaluation expression related to the one topic;
The conflict output expression output unit includes the conflict evaluation expression output step for outputting both a positive evaluation expression and a negative evaluation expression related to the one topic acquired in the conflict evaluation expression acquisition step. A method ,
In the recording medium,
One or more words included in the evaluation expression, the original form of each word, the part of speech of each word, the evaluation polarity of each word, and whether the word is a starting word or a word in the middle of the evaluation expression Learning data having type information indicating the type of each of the phrases indicating whether the phrase is irrelevant to the evaluation expression, and information relating to an arrangement of one or more words included in the evaluation expression, and learning data for each evaluation expression Two or more learning data storage units that can be stored,
The evaluation expression extraction step includes:
Morphological analysis of each sentence constituting the one or more sentences to obtain one or more words / phrases,
From the processing result in the morphological analysis sub-step, the one or more words / phrases acquired in the morphological analysis sub-step, the original form of the words / phrases, the part of speech of the words / phrases, and the arrangement of one or more words / phrases included in the evaluation expression A feature acquisition sub-step of acquiring information, searching the evaluation phrase dictionary using one or more words as a key, and acquiring evaluation polarities of the one or more words for each sentence;
One or more columns of type information indicating the types of phrases constituting each sentence by machine learning using the information acquired in the feature acquisition substep as features and using two or more learning data in the learning data storage unit A machine learning substep to obtain information about,
Of the one or more types of information columns acquired in the machine learning sub-step, one or more types corresponding to a type information sequence in which type information indicating a start phrase of evaluation expression or a phrase in the middle of evaluation expression is continuous An expression output method comprising: an evaluation expression acquisition sub-step of acquiring an evaluation expression that is a set of consecutive words . On the recording medium,
One or more sentences,
Storing one or more evaluation word / phrase information having an evaluation word / phrase relating to evaluation and an evaluation polarity which is information indicating whether the evaluation word / phrase is positive or negative;
An expression output method realized by an evaluation expression extraction unit, an evaluation expression evaluation expression clustering unit, an opponent evaluation expression acquisition unit, and an opponent evaluation expression output unit,
The evaluation expression extraction unit is a sentence that constitutes a sentence stored in the recording medium, and includes the evaluation phrase using an evaluation phrase stored in the recording medium from a sentence related to the topic An evaluation expression extraction step of extracting two or more evaluation expressions that are a set of one or more words;
The evaluation expression evaluation expression clustering unit clusters the two or more evaluation expressions from one or more words / phrases included in each of the two or more evaluation expressions extracted in the evaluation expression extraction step. An evaluation expression evaluation expression clustering step for obtaining one or more evaluation expression groups including:
The conflict evaluation expression acquisition unit uses an evaluation word and an evaluation polarity included in the recording medium from two or more evaluation expressions included in any evaluation expression group of the one or more evaluation expression groups, An opposing evaluation expression acquisition step of acquiring both a positive evaluation expression and a negative evaluation expression related to the one topic;
The conflict output expression output unit includes the conflict evaluation expression output step for outputting both a positive evaluation expression and a negative evaluation expression related to the one topic acquired in the conflict evaluation expression acquisition step. A method,
The recording medium is
An evaluation polarity string that is a string of evaluation polarities composed of evaluation expressions, an evaluation polarity string that is an array of evaluation polarities corresponding to one or more words included in the evaluation expression, and 1 included in the evaluation expression An evaluation expression that associates each of the above phrases, the original form of each of the phrases, the part of speech of each of the phrases, and an evaluation expression polarity that is information indicating whether the evaluation expression is positive or negative Further comprising evaluation expression learning data storage means capable of storing two or more learning data;
The conflict evaluation expression acquisition step includes:
Two or more evaluation expressions included in any one of the one or more evaluation expression groups are divided into words, one or more words are obtained, and the original form of each of the one or more words and A phrase information acquisition substep for acquiring a part of speech;
Using the one or more words acquired in the word information acquisition sub-step as a key, the evaluation word dictionary is searched, the evaluation polarity corresponding to each word is acquired, and the evaluation polarity that is a column of the one or more evaluation polarities An evaluation polarity column acquisition substep for acquiring columns for each evaluation expression;
One or more words / phrases acquired in the word / phrase information acquisition sub-step, the original form of each word, the part of speech of each word, and the evaluation polarity string acquired by the evaluation polarity string acquisition means are stored in the evaluation expression learning data storage means An evaluation expression polarity acquisition sub-step of applying to each of the two or more evaluation expression learning data being performed using machine learning and acquiring the evaluation expression polarity of each evaluation expression for each evaluation expression;
Opposite evaluation to acquire an evaluation expression in which the evaluation expression polarity acquired in the evaluation expression polarity acquisition substep is positive and an evaluation expression in which the evaluation expression polarity acquired in the evaluation expression polarity acquisition substep is negative An expression output method comprising an expression acquisition substep. On the recording medium,
One or more sentences,
Storing one or more evaluation word / phrase information having an evaluation word / phrase relating to evaluation and an evaluation polarity which is information indicating whether the evaluation word / phrase is positive or negative;
Computer
A sentence constituting a sentence stored in the recording medium, and a set of one or more words including the evaluation word / phrase using an evaluation word / phrase stored in the recording medium from a sentence related to the topic An evaluation expression extraction unit for extracting two or more evaluation expressions;
The two or more evaluation expressions are clustered from one or more words included in each of the two or more evaluation expressions extracted by the evaluation expression extraction unit, and one or more evaluation expression groups including one or more evaluation expressions are obtained. An evaluation expression clustering unit for
Using one or more evaluation expressions included in any one of the one or more evaluation expression groups, an evaluation phrase included in the recording medium and an evaluation polarity are used to positively determine the one topic. An opposing evaluation expression acquisition unit that acquires both evaluation expressions and negative evaluation expressions;
A program for functioning as a conflict evaluation expression output unit that outputs both a positive evaluation expression and a negative evaluation expression related to the one topic acquired by the conflict evaluation expression acquisition unit ,
The recording medium is
One or more words included in the evaluation expression, the original form of each word, the part of speech of each word, the evaluation polarity of each word, and whether the word is a starting word or a word in the middle of the evaluation expression Learning data having type information indicating the type of each of the phrases indicating whether the phrase is irrelevant to the evaluation expression, and information relating to an arrangement of one or more words included in the evaluation expression, and learning data for each evaluation expression Two or more learning data storage units that can be stored,
The evaluation expression extraction unit includes:
Morphological analysis of each sentence constituting the one or more sentences and obtaining one or more words;
From the processing result of the morpheme analysis unit, obtain information on one or more words acquired by the morpheme analysis unit, the original form of each of the words, the part of speech of each of the words, and the arrangement of one or more words included in the evaluation expression And a feature acquisition unit that searches the evaluation word dictionary using one or more words as a key, and acquires an evaluation polarity of each of the one or more words for each sentence;
Information on one or more columns of type information indicating the types of words constituting each sentence by machine learning using information acquired by the feature acquisition means as features and using two or more learning data in the learning data storage unit Machine learning means for acquiring
Among the one or more types of information column acquired by the machine learning means, one or more continuous types corresponding to the type information sequence in which the type information indicating that the evaluation expression is a starting word or phrase in the middle of the evaluation expression is continuous. A program for causing a computer to function as an evaluation expression acquisition unit that acquires an evaluation expression that is a set of phrases . On the recording medium,
One or more sentences,
Storing one or more evaluation word / phrase information having an evaluation word / phrase relating to evaluation and an evaluation polarity which is information indicating whether the evaluation word / phrase is positive or negative;
Computer
A sentence constituting a sentence stored in the recording medium, and a set of one or more words including the evaluation word / phrase using an evaluation word / phrase stored in the recording medium from a sentence related to the topic An evaluation expression extraction unit for extracting two or more evaluation expressions;
The two or more evaluation expressions are clustered from one or more words included in each of the two or more evaluation expressions extracted by the evaluation expression extraction unit, and one or more evaluation expression groups including one or more evaluation expressions are obtained. An evaluation expression clustering unit for
Using one or more evaluation expressions included in any one of the one or more evaluation expression groups, an evaluation phrase included in the recording medium and an evaluation polarity are used to positively determine the one topic. An opposing evaluation expression acquisition unit that acquires both evaluation expressions and negative evaluation expressions;
A program for functioning as a conflict evaluation expression output unit that outputs both a positive evaluation expression and a negative evaluation expression related to the one topic acquired by the conflict evaluation expression acquisition unit,
The recording medium is
An evaluation polarity string that is a string of evaluation polarities composed of evaluation expressions, an evaluation polarity string that is an array of evaluation polarities corresponding to one or more words included in the evaluation expression, and 1 included in the evaluation expression An evaluation expression that associates each of the above phrases, the original form of each of the phrases, the part of speech of each of the phrases, and an evaluation expression polarity that is information indicating whether the evaluation expression is positive or negative Further comprising evaluation expression learning data storage means capable of storing two or more learning data;
The conflict evaluation expression acquisition unit
Two or more evaluation expressions included in any one of the one or more evaluation expression groups are divided into words, one or more words are obtained, and the original form of each of the one or more words and Phrase information acquisition means for acquiring a part of speech;
Using the one or more words acquired by the word information acquisition means as a key, the evaluation word dictionary is searched, an evaluation polarity corresponding to each word is acquired, and an evaluation polarity column that is a column of the one or more evaluation polarities is obtained. Evaluation polarity string acquisition means for acquiring for each evaluation expression;
One or more each phrase acquired by the phrase information acquisition unit, the original form of each phrase, the part of speech of each phrase, and the evaluation polarity string acquired by the evaluation polarity string acquisition unit are stored in the evaluation expression learning data storage unit. An evaluation expression polarity acquisition unit that applies machine learning to two or more evaluation expression learning data, and acquires the evaluation expression polarity of each evaluation expression for each evaluation expression;
Opposite evaluation expression acquisition means for acquiring an evaluation expression in which the evaluation expression polarity acquired by the evaluation expression polarity acquisition means is positive and an evaluation expression in which the evaluation expression polarity acquired by the evaluation expression polarity acquisition means is negative; A program for causing a computer to function.

Images