JP5130900B2 - Opinion extraction device - Google Patents

Description

  The present invention relates to a technique for extracting a major (majority) opinion and a minor (minority) opinion from opinions received from users.

  In recent years, various services have been provided on the Internet. As one example, there is a Web site (hereinafter referred to as an evaluation site) that collects and publishes evaluations and opinions by general users regarding products, services, and other events. In such an evaluation site, when a product name or the like is input, evaluation information and opinion information about the product by a general user are displayed. For this reason, the use of the evaluation site has an advantage that it is possible to know the evaluation and opinion of the user who actually uses the product, not the information close to the seller of the product.

In a conventional evaluation site, it is possible to know how many good evaluations and bad evaluations exist for a certain event or the like, and it is also possible to know individual opinions. However, with regard to individual opinions, it is only possible to browse the ones selected arbitrarily from the received opinions. For example, if there are many good evaluations about an event, what points should be noted There is a problem that it is difficult to know a specific opinion whether many people say it is good. On the other hand, even if there are minority opinions, there are also valuable opinions, but here there is a problem that it is difficult to extract because there are few opinions. Although there is a technique for collecting a minority opinion with respect to a question such as a questionnaire (see Patent Document 1), it does not extract both a majority opinion and a minority opinion.
JP 2005-92616 A

  Therefore, an object of the present invention is to provide an opinion extraction device capable of extracting a majority opinion or a minority but valuable opinion.

  In order to solve the above problems, according to the present invention, article data storage means storing article data, evaluation information specifying means for specifying a combination of an evaluation object and an evaluation expression, and a combination of the specified evaluation object and evaluation expression , Target explanation word extracting means for extracting a target explanation word for explaining an evaluation target from article data stored in the article data storage means, and appearance count counting means for counting the number of appearances of the extracted target explanation word , A majority opinion extracting means for extracting article data having a target explanatory word with a large number of appearances as a result of the counting, and a sentence having a target explanatory word with a low number of appearances as a target explanatory text as a result of the counting A target explanation sentence extracting means, and the number of characters of the extracted target explanation sentence, or a target explanation sentence having a large number of predetermined words in the target explanation sentence as a minority opinion Providing an opinion extraction device having a minority opinion extracting means for extracting Te.

  According to the present invention, it is possible to extract a majority opinion or a minority but valuable opinion from among opinions received for a certain event.

(1. First embodiment)
DESCRIPTION OF EXEMPLARY EMBODIMENTS Hereinafter, preferred embodiments of the invention will be described in detail with reference to the drawings. FIG. 1 is a configuration diagram of an opinion extraction apparatus according to the first embodiment of the present invention. In FIG. 1, 10 is an article data storage means, 20 is an evaluation information designation means, 30 is an object explanation word extraction means, 40 is an appearance count counting means, 50 is a multiple opinion extraction means, 60 is an object explanation sentence extraction means, and 70 is Minority opinion extraction means, 80 is opinion presentation means, and 90 is arithmetic processing means.

  The article data storage means 10 stores article data in text format and article IDs for specifying article data in association with each other, and is realized by a storage device such as a hard disk connected to a computer. The evaluation information designating unit 20 designates a combination of an evaluation object and an evaluation expression, and is realized by an instruction input device such as a mouse and a keyboard, an input device I / F, and an OS function. The arithmetic processing unit 90 is realized by a CPU and a memory of a computer. When the CPU reads a dedicated program into the memory and executes it, the target explanatory word extraction unit 30, the appearance count counting unit 40, the majority opinion extraction unit 50, the target Functions as the explanatory note extracting unit 60 and the minority opinion extracting unit 70 are realized.

  The target explanatory word extraction unit 30 searches the article data storage unit 10 for a sentence having the specified evaluation target and evaluation expression, and extracts a target explanatory word that is a word explaining the specified evaluation target. The appearance count counting unit 40 counts how many times each target explanatory word extracted by the target explanatory word extraction unit 30 appears in all article data in the article data storage unit 10. The majority opinion extracting unit 50 identifies a predetermined number of target explanatory words having a large number of appearances among the target explanatory words counted by the appearance number counting unit 40, and converts article data including the specified target explanatory words into article data. Extracted from the storage means 10. The target explanatory sentence extracting unit 60 extracts a sentence including a predetermined number of target explanatory words with a small number of appearances as the target explanatory sentence from the target explanatory words counted by the appearance number counting unit 40. The minority opinion extraction unit 70 extracts, from the article data storage unit 10, as a minority opinion, the target explanatory text extracted by the target explanatory text extraction unit 60 with a large number of characters or a predetermined number of words.

  The opinion presentation unit 80 presents article data extracted by the majority opinion extraction unit 50 and the minority opinion extraction unit 70, and is realized by a display device such as a liquid crystal display, an output I / F, and an OS function. .

  Here, information stored in the article data storage unit 10 will be described. FIG. 2 is a diagram illustrating an example of information stored in the article data storage unit 10. As shown in FIG. 2, the article data storage means 10 stores the contents of article data specified by each article ID in association with the article ID. The article data stored in the article data storage means 10 is actually input from a user who has accessed the evaluation site published on the Internet, and the server computer that has established the evaluation site The article data transmitted from the user is registered in the article data storage means 10. Also, there are cases where the description content on the blog or bulletin board on the Web is acquired as article data and registered in the article data storage means 10. Further, the questionnaire data may be digitized and registered in the article data storage means 10 as article data.

  Next, the processing operation of the opinion extraction apparatus shown in FIG. 1 will be described. First, the user designates an evaluation object and an evaluation expression from the evaluation information designating unit 20. Then, the target explanatory word extraction unit 30 searches the article data storage unit 10 in full text with the specified evaluation target and evaluation expression, and extracts article data having both the specified evaluation target and evaluation expression. For example, when there are five articles as shown in FIG. 2, it is assumed that “information” is designated as the evaluation target and “want” is designated as the evaluation expression. In this case, articles K0001, K0002, and K0003 having both “information” and “want” are extracted.

  Subsequently, the target explanatory word extraction unit 30 performs syntax analysis of the extracted article. Then, as a result of the syntax analysis, a process is performed in which a specific part of speech is identified as a target explanatory word that is an evaluation target explanatory word among words related to the evaluation target. The part of speech to be the target explanatory word is set in advance, and the target explanatory word extraction means 30 extracts the word of the set part of speech as the target explanatory word. The part of speech that should be the target explanatory word can be set as appropriate, but in the present embodiment, a general noun, a noun that becomes a sub-variant connection, an independent adjective, and the like are set. In the example shown in FIG. 2, “player”, “injury”, “fashion”, “private”, and “game bulletin” are extracted as target explanatory words.

  Next, the appearance count counting unit 40 counts the number of times the target explanatory word extracted by the target explanatory word extraction unit 30 appears as the target explanatory word in the article data storage unit 10 for each target explanatory word. Therefore, even if the word is the same as the target explanatory word, it is not counted unless it appears as the target explanatory word. In the example shown in FIG. 2, the target explanation word “player” is twice, the target explanation word “injury” is once, the target explanation word “fashion” is once, the target explanation word “private” is once, the target explanation The word “game bulletin” will be counted once. For counting the number of appearances, in addition to the method of simply counting in this way, a method of counting by a combination of target explanatory words may be used. For example, in the case of the article K0002 shown in FIG. 2, the evaluation target “information” appears twice, but there are two target explanation words for “information” in the above “information” and “fashion”. Since all are related to the same evaluation object “information”, the combination of “player” and “fashion” is counted as one time.

  Next, the majority opinion extraction means 50 selects a predetermined number of target explanatory words with a large number of appearances counted by the appearance count counting means 40, and selects a predetermined number from those with a large number of appearances. Extract as As described above, article data that is a majority opinion is extracted from the article data storage means 10 by the target explanatory word extraction means 30, the appearance count counting means 40, and the majority opinion extraction means 50.

  Opinion extraction apparatus 100 extracts minority opinions in parallel with majority opinions. When the evaluation target and the evaluation expression are specified, the processing in the target explanatory word extraction unit 30 and the appearance count counting unit 40 is the same as the case of extracting a majority opinion, and the results up to this point are used together with the majority opinion extraction. . Then, the minority opinion extraction unit 70 selects a predetermined number of target explanatory words with a small number of appearances counted by the appearance number counting unit 40 from those with a small number of appearances. Target explanatory words with a small number of appearances often have the same number of appearances, and are difficult to select as they are. For this reason, further narrowing processing is performed. Specifically, first, the target explanatory sentence extracting means 60 extracts a sentence including the target explanatory word selected as the target explanatory word as the target explanatory word from the article data as the target explanatory sentence. The sentence is specified by a known parsing technique. Then, in the extracted target explanatory text, the target explanatory text having a large number of characters or a predetermined number of words is extracted as a minority opinion. For example, as a thing with a large number of characters, the extracted object explanation sentence having a long length, that is, an object explanation sentence with a large number of characters is extracted as a minority opinion. In addition, there are a method for extracting a large number of dependency words and a method for extracting a large number of target explanatory words as a method having a large number of predetermined words. In this way, minority opinions are extracted from the article data storage means 10 by the target explanatory word extraction means 30, the appearance count counting means 40, the majority opinion extraction means 50, the target explanation sentence extraction means 60, and the minority opinion extraction means 70.

  When the majority opinion is extracted by the majority opinion extraction means 50 and the minority opinion is extracted by the minority opinion extraction means 70, the opinion presentation means 80 creates a title including the target explanatory word and the evaluation object for each of the majority opinion and the minority opinion. To do. The title of the majority opinion is created by adding an appropriate particle between the target explanatory word and the evaluation target. For example, when the target explanatory word is “player” and the evaluation target is “information”, a title “player information” is created. The title of the minority opinion is created by extracting all character strings between the target explanatory words and the evaluation target from the article data. Then, screen data is created and presented by arranging the created titles in a layout set in advance for each of the majority opinion and the minority opinion.

  The state of the screen presented at this time is shown in FIG. In FIG. 3, the evaluation target and the evaluation expression designated by the evaluation information designating unit 20 are displayed at the top. A list of “majority comments” is displayed at the top of the screen, a “content of opinion” is displayed at the center of the screen, and a list of “minority comments” is displayed at the bottom of the screen. In the majority comment column, the titles of articles extracted as majority comments are displayed in a list. When an instruction is given to select each title, the content of the article is changed to “content of opinion” and “content of opinion”. "" Is displayed. The example of FIG. 3 shows a case where “player information” is selected. In the minority comments column, the titles of articles extracted as minority comments are listed.

  The opinion extraction apparatus according to the first embodiment can be used as a stand-alone type or a network type. For example, when questionnaire results are collected as article data and analyzed by a single computer to extract majority opinions and minority opinions, it is realized as a stand-alone type. Moreover, in order to allow general users to browse from their own terminals, a part of the evaluation information specifying means 20 and the opinion presentation means 80 is configured with a network communication function, thereby realizing a network type.

(2. Second Embodiment)
Next, a second embodiment will be described. FIG. 4 is a configuration diagram of an opinion extraction apparatus according to the second embodiment of the present invention. In FIG. 4, 11 is an access log storage means, 71 is a minority opinion extraction means, 85 is an access log acquisition means, and 91 is an arithmetic processing means. In FIG. 4, the same components as those in FIG.

  The access log storage unit 11 stores an access log to the article data stored in the article data storage unit 10, and is realized by a storage device such as a hard disk connected to the computer, like the article data storage unit 10. The The opinion extraction apparatus 101 according to the second embodiment has a function as a Web server, and presents article data stored in the article data storage unit 10 to a terminal that has accessed through a network. When the article data is read from the terminal, the user ID of the user is acquired, and the user ID and the article ID are associated with each other and recorded in the access log storage unit 11 as a reading log. When writing is performed for the presented article, the user ID and the article ID are associated with each other and recorded in the access log storage unit 11 as a writing log. In this specification, the read log and the write log are collectively referred to as an access log. For example, the user ID can be input every time when logging in to the opinion extraction device, the cookie file is used, the user ID is assigned to the acquired IP address, and the ID corresponding to the IP address is used from the next time on. It can be obtained by a known method such as obtaining.

  An example of information stored in the access log storage unit 11 is shown in FIG. FIG. 5A shows a write log, and FIG. 5B shows a read log. As shown in FIGS. 5A and 5B, the access log storage unit 11 stores article IDs that are written or read in association with user IDs.

  The access log acquisition unit 85 acquires the access log stored in the access log storage unit 11 and passes it to the minority opinion extraction unit 71. The minority opinion extraction means 71 specifies a lower predetermined number of target explanation words with a small number of appearances among the target explanation words counted by the appearance count counting means 40, and selects article data including the specified target explanation words. Then, article data having a high score calculated using the access log is extracted from the article data storage means 10 among the selection candidates.

  Next, the processing operation of the opinion extraction apparatus shown in FIG. 4 will be described. The extraction of the majority opinion is exactly the same as the processing in the first embodiment. Regarding the extraction of minority opinions, the processing in the target explanatory word extraction means 30 and the appearance count counting means 40 is the same as the case of extracting a majority opinion. Then, the minority opinion extracting unit 71 selects a predetermined number of target explanatory words with a small number of appearances counted by the appearance number counting unit 40. Specifically, first, the target explanatory sentence extracting means 60 extracts an article including the target explanatory word selected as the target explanatory word as the target explanatory word as the target explanatory sentence.

  Further, the minority opinion extracting unit 71 acquires an access log for the target explanatory text extracted by the target explanatory text extracting unit 60 from the access log storage unit 11 via the access log acquiring unit 85. Then, the score of each article ID for writing and reading is calculated using the writing log and the reading log, respectively. The score is calculated by increasing the score of an article ID written or read by a user who writes and reads a lot.

  For example, consider the case where the score of each article ID is calculated using the write log of FIG. In the example of FIG. 5A, the user U0001 has written three articles K0011, K0014, and K0015. In this case, the score “3” is given to each of the articles K0011, K0014, and K0015. Users U0002 and U0003 have written only one of articles K0012 and K0013, respectively. Therefore, each of the articles K0012 and K0013 is given a score “1”.

  Also, consider a case where the score of each article ID is calculated using the read log of FIG. In the example of FIG. 5B, the user U0001 reads two articles K0011 and K0012. In this case, each of the articles K0011 and K0012 is given a score “2”. User U0002 has read only one article K0012. Therefore, the score “1” is given to the article K0012. The article K0012 is given a score “2” when focusing on the user U0001, and a score “1” when focusing on the user U0002. In this case, the lower score “1” is given to the article K0012. Also, K0013, K0014, and K0015 are not read and are given a score of “0”.

  Further, the minority opinion extraction unit 71 calculates a total score using the writing score and the reading score. In the present embodiment, the minority opinion extraction unit 71 performs the calculation according to the following [Formula 1].

[Formula 1]
(Overall score) = α × (Write score) + β × (Reading score)

  In [Equation 1], α and β are coefficients, and can be set as appropriate. When α = β = 1, in the example of FIGS. 5A and 5B, the overall score of the article K0011 is the largest at 5, so the minority opinion extraction means 71 determines the article K0011 as a minority opinion, Extract.

  When the majority opinion is extracted by the majority opinion extraction means 50 and the minority opinion is extracted by the minority opinion extraction means 71, the opinion presentation means 80, for each of the majority opinion and the minority opinion, is the target explanation word as in the first embodiment. And the character string between the evaluation targets is extracted from the article as the title. Then, the title is arranged for each of the majority opinion and the minority opinion in a preset layout, and screen data is created and presented. If the presentation destination is a terminal via the Internet, HTML data is created and transmitted to the terminal that is the access source.

It is a lineblock diagram of an opinion extraction device concerning a 1st embodiment of the present invention. 4 is a diagram illustrating an example of information stored in an article data storage unit 10. FIG. It is a figure which shows the mode of the screen shown by the presentation means. It is a block diagram of the opinion extraction device which concerns on the 2nd Embodiment of this invention. It is a figure which shows an example of the information memorize | stored in the access log memory | storage means.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 ... Article data storage means 11 ... Access log storage means 20 ... Evaluation information designation means 30 ... Target explanatory word extraction means 40 ... Appearance count counting means 50 ... Multiple opinion extraction means 60 ... Object explanation sentence extraction means 70, 71... Minority opinion extraction means 80... Opinion presentation means 85... Access log acquisition means 90.

Claims (4)

Article data storage means for storing article data;
An evaluation information specifying means for specifying a combination of an evaluation object and an evaluation expression;
A target explanatory word extracting means for extracting a target explanatory word for explaining the evaluation target from the article data stored in the article data storage means according to the combination of the designated evaluation target and the evaluation expression;
Appearance count counting means for counting the number of appearances of the extracted target explanatory words;
As a result of the counting, a majority opinion extracting means for extracting article data having a target explanatory word with a large number of appearances as a majority opinion;
As a result of the counting, target explanation sentence extracting means for extracting a sentence having a target explanation word with a small number of appearances as a target explanation sentence;
A minority opinion extracting means for extracting the number of characters of the extracted target explanatory text or the target explanatory text having a large number of predetermined words in the target explanatory text as a minority opinion;
An opinion extraction device characterized by comprising: Access log storage means for storing an access log to the article data;
Access log acquisition means for acquiring an access log related to a predetermined article from the access log storage means,
The minority opinion extraction means uses the acquired access log from among the article data having the extracted target explanatory text, and gives a higher score to the target explanatory text included in the article data accessed by a user with much access. 2. The opinion extraction apparatus according to claim 1, wherein a predetermined number of target explanatory sentences having a high score are extracted as minority opinions. The access log consists of a read log and a write log,
The score is higher for a target explanatory sentence included in article data read by a user who reads a lot, and higher for a target explanatory sentence included in article data written by a user who writes a lot. The opinion extraction device according to claim 2, wherein the opinion extraction device is a feature.   The program for functioning a computer as an opinion extraction apparatus in any one of Claims 1-3.

Images