JP5138622B2 - Information processing apparatus, complaint extraction method, and program - Google Patents

Description

  The present invention relates to an information processing apparatus, a dissatisfaction extraction method, and a program, and in particular, extracts “dissatisfaction” from a plurality of documents (information such as impressions and experiences related to products and services transmitted by a large number of consumers). The present invention relates to an information processing apparatus, a dissatisfaction extraction method, and a program.

  In recent years, in the information that consumers send on the Internet, impressions about products and services, daily experiences, etc. have been sent in large quantities. Under such circumstances, assuming a situation where users search for evaluations of certain things from information sent by a large number of consumers, evaluation attributes such as “screen” and “pretty” Screen) and evaluation expression (beautiful) are extracted (see, for example, Patent Document 1). Specifically, the reputation related to the search word is extracted by looking at the similarity between the candidates for the reputation expression extracted from the document group including the search word. The reputation extracted in this way includes complaints such as “dirty screen”.

JP 2008-234557 A

  When a user's dissatisfaction is extracted using a conventional reputation extraction technique, the main use is information collection such as marketing, and it is difficult to apply it to product recommendation. The reason for this is that there are various types of dissatisfaction, such as criticism, feelings, demerits, and detrimental events. It is mentioned that necessary things are included.

  The present invention has been made in view of the above points, and is an information processing apparatus capable of identifying and extracting dissatisfactions regarding disadvantages and disadvantageous events as dissatisfactions that can be used for other services from various dissatisfactions. It is another object of the present invention to provide a complaint extraction method and program.

  FIG. 1 is a principle configuration diagram of the present invention.

The present invention (Claim 1) is an information processing apparatus for extracting dissatisfaction,
An improvement operation extraction unit 110 that acquires a document from the document set database 10 holding a document group, extracts an improvement operation, and stores it in the improvement operation storage unit 20;
An operation target extraction unit 120 that acquires an improvement operation from the improvement operation storage unit 20, extracts a target of the improvement operation from the document acquired from the document set database 10, and obtains a co-occurrence frequency of the improvement operation and the target of the improvement operation. When,
The improvement operation acquired from the operation target extraction unit 120 and the co-occurrence frequency of the improvement operation target are acquired, the improvement operation target is classified into one of negative, neutral, and positive evaluation polarities, and the improvement operation target Polarity determination means 130 for outputting a polarity label;
And a dissatisfaction output means 140 that obtains the object of the improvement operation and the polarity label and outputs the object of the improvement operation having a negative polarity label as dissatisfaction.

Further, the present invention (Claim 2) obtains dissatisfaction from the dissatisfaction output means 140, and a word that matches the dissatisfaction from the user text storage means for storing the text described by the user who recommends the page of the Q & A site. A solution object dissatisfaction extraction means for extracting
A question search means for obtaining a question posted by the user from the Q & A site storage means, and selecting a question including a word that is unsatisfactory from the questions;
It further includes recommendation page output means for acquiring a question selected by the question search means, and acquiring and outputting an answer corresponding to the question from the Q & A site storage means.

Further, the present invention (Claim 3) includes a reputation extraction means for extracting a set of {attribute expression, evaluation expression, evaluation polarity} as a reputation from the document set database 10,
The classification result of the improvement operation target and the evaluation polarity is acquired from the polarity determination unit 130, the reputation is acquired from the reputation extraction unit, and the improvement operation target highly related to the improvement operation is included in the attribute expression from the reputation. Reputation selection means to extract reputation and extract reputation of negative evaluation polarity;
Further comprising
The dissatisfied output means 140 is
Dissatisfaction is output based on the object of improvement action with negative polarity label and reputation of negative evaluation polarity.

  FIG. 2 is a diagram for explaining the principle of the present invention.

The present invention (Claim 4) is a dissatisfaction extraction method for extracting dissatisfaction,
An improvement operation extracting step (step 1) in which the improvement operation extracting means acquires a document from a document set database holding a document group, extracts the improvement operation, and stores it in the improvement operation storage unit;
An operation in which the operation target extraction unit acquires the improvement operation from the improvement operation storage unit, extracts the target of the improvement operation from the document acquired from the document set database, and obtains the co-occurrence frequency of the improvement operation and the target of the improvement operation An object extraction step (step 2);
The polarity determining means obtains the co-occurrence frequency of the improvement operation and the target of the improvement operation, classifies the target of the improvement operation as one of negative, neutral, and positive evaluation polarities, and sets the target of the improvement operation and the polarity label. Output polarity determination step (step 3);
The dissatisfied output means performs a dissatisfied output step (step 4) of acquiring the object of improvement operation and the polarity label, and outputting the object of improvement operation having a negative polarity label as dissatisfaction.

Further, according to the present invention (claim 5), the solution object dissatisfaction extracting means acquires the dissatisfaction and matches the dissatisfaction from the user text storage means for storing the text described by the user who is to recommend the page of the Q & A site. A dissatisfaction target extraction step for extracting words to be
A question search step in which the question search means obtains a question posted by the user from the Q & A site storage means, and selects a question including a dissatisfied word from the question;
The recommended page output means further performs a recommended page output step of acquiring a question including a word that is unsatisfactory and acquiring and outputting an answer corresponding to the question from the Q & A site storage means.

According to the present invention (claim 6), a reputation extraction step in which the reputation extraction means extracts a set of {attribute expression, evaluation expression, evaluation polarity} as a reputation from the document set database;
The classification result of the improvement action target and the evaluation polarity and the reputation are acquired, and the reputation having the object of the improvement action highly relevant to the improvement action in the attribute expression is extracted from the reputation, and the negative evaluation polarity Further performing a reputation selection step of extracting the reputation of
In the dissatisfied output step,
Dissatisfaction is output based on the object of improvement action with negative polarity label and reputation of negative evaluation polarity.

  The present invention (Claim 7) is a dissatisfaction extraction program for causing a computer to function as each means constituting the information processing apparatus according to any one of Claims 1 to 3.

  According to the present invention, it is possible to identify and extract dissatisfactions that can be used for services other than information collection from among dissatisfactions that are mixed and mixed. The extracted dissatisfaction is, for example, a service that automatically leads to a related Q & A site (a site where another user answers a question written by the user) in response to the dissatisfaction that the user has. It is effective against this, and is expected to lead to the resolution of user dissatisfaction and the activation of Q & A services.

It is a principle block diagram of this invention. It is a figure for demonstrating the principle of this invention. It is a flowchart of the dissatisfaction extraction procedure in one embodiment of this invention. It is a block diagram of the dissatisfaction extraction apparatus in 1st Example of this invention. It is an example of an output in each processing of the 1st example of the present invention. The extraction example of the improvement operation | movement of 1st Example of this invention is shown. It is a block diagram of the dissatisfaction extraction apparatus in 2nd Example of this invention. It is an example of an output in each processing of the 2nd example of the present invention.

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

  First, processing for solving the above-described problem will be described.

  FIG. 3 is a flowchart of the dissatisfaction extraction procedure in one embodiment of the present invention.

  In the first place, “dissatisfaction” regarding disadvantages and disadvantageous events is an object to be solved by the product, and is often described when explaining the effect of the product. As an example, there is a description of “I will eliminate lack of exercise”. At this time, the explanatory text is composed of dissatisfaction of “insufficient exercise” and an operation of “resolution”. Therefore, in order to solve the above-described problems, first, operations that are highly profitable for the receiver of the operation, such as “elimination”, “improvement”, and “prevention” from the document group of the document collection DB (hereinafter referred to as “improvement operation”). Is called (step 101).

  Next, the target of the improvement operation is extracted from the document group (step 102). Here, the “object of improvement operation” corresponds to “insufficient exercise” in the example of “elimination of lack of exercise”. However, since the object itself may not be dissatisfied as described in the description of “improve speed”, the polarity of the extracted object (positive / neutral / negative) is determined, and a negative object is extracted ( Step 103). This negative object becomes “dissatisfaction” with regard to disadvantages and disadvantageous events.

  Embodiments of the present invention will be described below with reference to the drawings.

[First embodiment]
As an example of the present invention, it is recommended to extract a dissatisfaction described in a text and recommend a page of a Q & A site (“Tell me! Goo” (registered trademark), etc.) asking about the same dissatisfaction as the extracted dissatisfaction. Think. As a result, it is expected that a text writer can easily obtain a solution to his dissatisfaction. For example, if the text contains the phrase “Lack of exercise recently,” the system will provide an answer page to the question “Are there any ways to resolve lack of exercise?” On the Q & A site. Recommend to text writers.

  FIG. 4 shows the configuration of the dissatisfaction extraction device in the first embodiment of the present invention.

  The dissatisfaction extraction apparatus shown in the figure includes a document set database (DB) 10, an improved operation list storage unit 20, a dissatisfaction list storage unit 30, a Q & A site database (DB) 40, a user text storage unit 50, an improved operation extraction unit 110, The operation target extraction unit 120, the polarity determination unit 130, the dissatisfaction output unit 140, the solution target dissatisfaction extraction unit 150, the question search unit 160, and the recommended page output unit 170 are configured.

  The document set DB 10 holds a plurality of digitized documents.

  The improved action list storage unit 20 holds the improved actions extracted by the improved action extractor 110.

  The dissatisfaction list storage unit 30 holds the dissatisfaction output from the dissatisfaction output unit 140.

  The Q & A site DB 40 holds a question posted on the Q & A site and an answer associated with the question.

  The user text storage unit 50 holds text written by a user who is a candidate for recommending a page of a Q & A site.

  Hereinafter, the operation of each component will be described.

  FIG. 5 shows an output example in each processing of the first embodiment of the present invention.

  The improvement operation extraction unit 110 extracts the improvement operation by the method shown in FIG. First, a document group held in the document set DB 10 is received. Next, we extract three objects (action shortage), action ("resolve"), and modality ("resolve me") from the received document (Fig. 6 (a)). The co-occurrence frequency A of the motion target and the co-occurrence frequency B of the modality representing the motion and profit are calculated (FIG. 6B). As a method for calculating the co-occurrence frequency, for example, morpheme analysis may be performed on all documents, and the number of occurrences of words whose co-occurrence frequencies are to be calculated within a certain number of morphemes may be counted. However, the co-occurring range is arbitrary, and the types of modalities representing the negative operation target and profit need not use all of the operation targets and modalities, and may be a small number of samples. Therefore, as a negative operation target, for example, a disease name or the like having a high appearance frequency in a document group may be extracted from a medical dictionary or the like, and a modality reference book (reference document 1) may be used for a modality representing profit. : The modality described as an example in Takashi Masuoka's “Exploration of Japanese Modality”, Kuroshio Publishing, 2007, etc.) may be used.

  Finally, the improvement action extraction unit 110 calculates a product of the co-occurrence frequency A and the co-occurrence frequency B including the same action (FIG. 6C), and arbitrarily selects an action with a high product value (improvement action) in descending order. N are extracted and stored in the improved operation list storage unit 20 (FIG. 5A).

  The operation target extraction unit 120 acquires the document group held in the document set DB 10 and the improvement operation held by the improvement operation list storage unit 20, extracts all the targets of the improvement operation from the document, and further, {improvement operation, improvement operation The co-occurrence frequency is calculated, and {target of improvement operation, co-occurrence frequency} is passed to the polarity determination unit 130 (FIG. 5B). However, the co-occurring range is arbitrary. In addition, regarding the method of extracting the target of the improvement operation, it is possible to use the existing morpheme analysis and syntax analysis to see the morpheme pattern, dependency, etc. immediately before the improvement operation.

  The polarity determination unit 130 acquires a document group from the document set DB 10 and acquires {target of improvement operation, co-occurrence frequency} from the operation target extraction unit 120. After that, the received improvement action targets are classified into three evaluation polarities: negative, neutral, and positive. The improvement action targets and polarity labels (negative, neutral, and positive) assigned to each target are assigned. It passes to the dissatisfied output part 140 (FIG.5 (c)). For classification of evaluation polarity, refer to Reference 2 “Higashiyama et al.,“ Acquiring Noun Evaluation Polarity Focusing on Preferential Predicate Preference ”, Proc. For the classifier created by the method described in the above, a method of extracting and inputting a predicate pattern associated with the object of the improvement operation from the document group is conceivable. In addition, regarding the extraction of the predicate pattern, the rules described in Reference Document 2 may be used. However, instead of passing all the improvement action targets to the dissatisfied output unit 140, only the improvement action target having a co-occurrence frequency higher than an arbitrary threshold value may be used. However, instead of passing all the improvement action targets to the dissatisfied output unit 140, only the improvement action target having a co-occurrence frequency higher than an arbitrary threshold value may be used.

  The dissatisfied output unit 140 receives the target of the improvement operation and the polarity label assigned to each target from the polarity determination unit 130. Next, the object of the improvement operation to which the negative polarity label is assigned is held in the dissatisfaction list storage unit 30 as “dissatisfied” (FIG. 5D).

  The solution target dissatisfaction extraction unit 150 acquires a dissatisfaction list from the dissatisfaction list storage unit 30, and receives text (FIG. 5 (e)) described by the user to be recommended from the user text storage unit 50. Then, words that match the dissatisfaction described in the received dissatisfaction list are extracted from the text described by the user and passed to the question search unit 160 (FIG. 5 (f)).

  The question search unit 160 receives the dissatisfaction described by the user to be recommended from the solution target dissatisfaction extraction unit 150 and the question posted to the Q & A site from the Q & A site DB 40. From the received questions, questions containing dissatisfaction described by the user are selected and all are sent to the recommended page output unit 170 (FIG. 5G).

  The recommendation page output unit 170 receives the question including the dissatisfaction described by the user from the question search unit 160 and the answer associated with the question from the Q & A site DB 40 and presents them to the user (FIG. 5H).

[Second Embodiment]
FIG. 7 shows the configuration of the dissatisfaction extraction device in the second embodiment of the present invention.

  The dissatisfaction extraction apparatus shown in the figure is configured by adding a reputation extraction unit 180 and a reputation selection unit 190 to the configuration of FIG. 4 of the first embodiment. An example of the output of each process is shown in FIG.

  In this embodiment, in addition to the “dissatisfaction” extracted in the first embodiment, complaints regarding disadvantages and disadvantageous events are extracted from the reputation extracted by the method described in Patent Document 1.

  Specifically, a reputation including an attribute expression that easily co-occurs with the improvement operation is extracted.

  As will be described in detail below, the improvement action extraction unit 110, the action target extraction part 120, the polarity determination part 130, the solution target dissatisfaction extraction part 150, the question search part 160, the recommended page output part 170, the document set DB 10, and the improvement action list storage Since the unit 20, the dissatisfaction list storage unit 30, the Q & A site DB 40, and the user text storage unit 50 are the same as those in the first embodiment, the description thereof is omitted.

  The reputation extraction unit 180 acquires a document group held in the document set DB 10. Next, the method described in Patent Document 1 and Reference Document 3 “Daiya Takamura, et al.,“ Modeling Evaluation Expressions Consisting of Multiple Words ”, Annual Conference of the Language Processing Society of Japan, Vol. 12th, Page 524-527, 2006 ”, a set of {attribute expression, evaluation expression, evaluation polarity} is extracted as a reputation from the document group and passed to the reputation selection unit 190.

  First, the reputation selection unit 190 receives the improvement operation target, the classification result of the evaluation polarity, and the co-occurrence frequency of the improvement operation from the polarity determination unit 130. Next, a set of {attribute expression, evaluation expression, evaluation polarity} is received as a reputation from the reputation extraction unit 180. The target of the improvement operation is likely to include an attribute expression that is a target that the user feels dissatisfied like “speed (improves)”. Therefore, from the reputation received from the reputation extraction unit 180, a reputation having, as attribute expressions, an object of the improvement operation highly related to the improvement operation is extracted. Regarding the high degree of relevance to the improvement operation, it is conceivable to use a statistical index such as the co-occurrence frequency of the improvement operation and the object of the improvement operation, the lift value, or the like. Finally, among the extracted reputations, a negative evaluation polarity reputation is passed to the dissatisfaction output unit 140.

  The dissatisfaction output unit 140 receives a reputation of negative evaluation polarity from the reputation selection unit 190, and holds it in the dissatisfaction list storage unit 30 as dissatisfaction together with the output of the dissatisfaction output unit 140 in the first embodiment.

  The subsequent processing is the same as in the first embodiment.

  In addition, the operation of the dissatisfaction extraction device shown in the first embodiment and the second embodiment is constructed as a program and installed and executed on a computer used as the dissatisfaction extraction device, or via a network. It can be distributed.

  Further, the constructed program can be stored in a portable storage medium such as a hard disk, a flexible disk, or a CD-ROM, and can be installed or distributed in a computer.

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

  The present invention can be automatically guided to a related Q & A site for the dissatisfaction that the user has, and is expected to lead to the user's dissatisfaction resolution, activation of the Q & A service, and the like.

10 document set database 20 improved action storage means, improved action list storage section 30 dissatisfaction list storage section 40 Q & A site database 50 user text storage section 110 improved action extraction means, improved action extraction section 120 action target extraction means, action target extraction section 130 Polarity determination unit, polarity determination unit 140 Dissatisfaction output unit, dissatisfaction output unit 150 Resolution target dissatisfaction extraction unit 160 Question search unit 170 Recommended page output unit 180 Reputation extraction unit 190 Reputation selection unit

Claims (7)

An information processing apparatus for extracting dissatisfaction,
Improvement operation extracting means for acquiring a document from a document set database holding a document group, extracting an improvement operation, and storing it in the improvement operation storage means;
An action target extracting means for obtaining an improvement action from the improvement action storage means, extracting a target of the improvement action from a document acquired from the document set database, and obtaining a co-occurrence frequency of the improvement action and the improvement action target; ,
Obtaining the co-occurrence frequency of the improvement action and the improvement action target acquired from the action target extracting means, classifying the improvement action target into one of negative, neutral, and affirmative evaluation polarities; And polarity determination means for outputting a polarity label;
A dissatisfaction output means for acquiring the improvement operation target and the polarity label, and outputting the improvement operation target having a negative polarity label as dissatisfaction;
An information processing apparatus comprising: The dissatisfaction extraction means for extracting the word that matches the dissatisfaction from the user text storage means for acquiring the dissatisfaction from the dissatisfaction output means and storing the text described by the user who is the target of recommending the page of the Q & A site;
A question search means for obtaining a question posted by a user from the Q & A site storage means, and selecting a question including a word that matches the dissatisfaction from the questions;
A recommendation page output means for acquiring the question selected by the question search means, acquiring and outputting an answer corresponding to the question from the Q & A site storage means,
The information processing apparatus according to claim 1, further comprising: Reputation extraction means for extracting a set of {attribute expression, evaluation expression, evaluation polarity} as a reputation from the document set database;
The improvement operation target and the evaluation polarity classification result are acquired from the polarity determination unit, the reputation is acquired from the reputation extraction unit, and the improvement operation target highly relevant to the improvement operation is attributed from the reputation Reputation selection means to extract the reputation of the expression and extract the reputation of negative evaluation polarity,
Further comprising
The dissatisfied output means includes
The information processing apparatus according to claim 1, wherein the information processing apparatus outputs dissatisfaction based on an object of the improvement operation having the negative polarity label and a reputation of the negative evaluation polarity. A complaint extraction method for extracting complaints,
An improvement action extracting means for acquiring a document from a document set database holding a document group, extracting an improvement action, and storing the extracted action in an improvement action storage means;
The action target extracting means acquires the improvement action from the improvement action storage means, extracts the improvement action target from the document acquired from the document set database, and determines the co-occurrence frequency of the improvement action and the improvement action target. A desired action extraction step;
The polarity determination means obtains the co-occurrence frequency of the improvement operation and the target of the improvement operation, classifies the target of the improvement operation as one of the negative, neutral, and positive evaluation polarities. Polarity determination step for outputting
A dissatisfied output means obtains the object of the improvement operation and the polarity label, and outputs the object of the improvement operation having a negative polarity label as dissatisfaction, and
Dissatisfaction extraction method characterized by performing. Solution target dissatisfaction extraction means for extracting the word that matches the dissatisfaction from the user text storage means for acquiring the dissatisfaction and storing the text described by the user who is the target of recommending the page of the Q & A site When,
A question search step in which the question search means acquires a question posted by the user from the Q & A site storage means and selects a question including a word that matches the dissatisfaction from the questions;
Recommendation page output means, to get the questions that contain the word that matches before Symbol complaints, and the recommendation page output step of outputting to get the answer corresponding to the question from the Q & A site storage means,
The dissatisfaction extraction method of Claim 4 which performs further. A reputation extraction step in which a reputation extraction means extracts a set of {attribute expression, evaluation expression, evaluation polarity} as a reputation from the document set database;
The classification result of the improvement action target and the evaluation polarity and the reputation are acquired, and the reputation having the attribute of the improvement action highly relevant to the improvement action is extracted from the reputation, and negative Further performing a reputation selection step of extracting reputation of evaluation polarity,
In the dissatisfied output step,
The dissatisfaction extraction method according to claim 4 or 5 , wherein dissatisfaction is output based on an object of the improvement operation having the negative polarity label and a reputation of the negative evaluation polarity. A dissatisfaction extraction program for causing a computer to function as each means constituting the information processing apparatus according to any one of claims 1 to 3.

Images