JP6048691B2 - Dissatisfaction extraction device, dissatisfaction extraction method, and dissatisfaction extraction program - Google Patents

Description

  Dissatisfaction extraction device, dissatisfaction extraction method, and program, which can extract dissatisfaction useful for improvement of evaluation target products from documents written by users of evaluation target products such as blogs and Internet bulletin boards The present invention relates to a dissatisfaction extraction apparatus, method, and program.

  In recent years, purchasers and users of products have been sending a large amount of facts and opinions regarding purchased products and used products on blogs and Internet bulletin boards. The opinions that buyers and users have about the product include positive opinions such as “Easy to use” and negative opinions (dissatisfaction) such as “Screen is dark and difficult to see” and “ There are various opinions such as "I want to increase the resolution". Some of these opinions are useful for improving products.

  As a useful opinion for such improvement, there is a direct request such as “I want to make it easier to press keys”. With regard to the direct request, the direct request can be extracted by using an expression representing the request such as “I want” as described in Non-Patent Document 1.

  On the other hand, even though dissatisfaction, because of the complaining level of “cell phone A cannot be used”, “cell phone A has function α, but cell phone B does not have function α” etc. Various complaints are included, up to those useful for improving the product.

  That is, in the complaint that “the mobile phone A cannot be used”, the product developer does not know which point to improve. On the other hand, from the dissatisfaction that “the mobile phone A has the function α but the mobile phone B does not have the function α”, the developer of the mobile phone B adds the function α to the mobile phone B. Can be considered. The function α is realized in the mobile phone A, and there is almost no technical problem in adding the function α to the mobile phone B. Such dissatisfaction is also an indirect request.

  In this way, if a dissatisfaction useful for improving the evaluation target product can be extracted, the product developer can immediately provide feedback to the product development.

  Although the user is not dissatisfied, the user of the mobile phone C is not dissatisfied, such as “the mobile phone C is also good, but the mobile phone D is even better”. You may know that it exists. In this case, since the user feels that the mobile phone D is superior to the mobile phone C and can be regarded as a potential dissatisfaction possessed by the user of the mobile phone C, such dissatisfaction also represents an expectation for improvement. As a result, if product feedback can be fed back to the product development, the product developer can develop the product in line with consumer expectations.

  Dissatisfaction that expresses expectations for improvement is defined as dissatisfaction that states that there is a better comparison target than the function, performance, and design of the evaluation target, and that the evaluation target is inferior to the comparison target. . Such a dissatisfaction is a dissatisfaction useful for improving the evaluation target product.

The following are examples of dissatisfaction that expresses expectations for improvement of the product to be evaluated and other dissatisfactions, and the reasons are shown below.
○ Example of dissatisfaction that expressed the expectation of improvement for the product to be evaluated:
“I can play music on my computer, but I can't play it when I import it into portable audio.”
"It is a pity that flick input is possible with the smartphone OS made by company G, but flick input is not possible with the smartphone OS made by company A."
"The LCD screen of the B company smartphone is beautiful, but the A company smartphone is dirty."
"The screen of the C company smartphone is beautiful, but the screen of the B company smartphone cannot be complaining." )
○ Non-satisfactory examples of expectations for improvement of the product being evaluated and the reasons:
“On the second day, the battery could not be charged normally and the red lamp flashed.” (Reason: only fact, not opinion)
“It's very unfortunate to have a swing setting that doesn't work.” (Reason: I don't know how to use it, so I can't give feedback on product development.)
“I wish it would be possible to auto-refresh the screen.” (Reason: This is a request and it is necessary to examine whether it is feasible or technical.)
“The screen of the D company smartphone is not good, but the screen of the E company smartphone is too harsh.” Even the user is not satisfied.)

  By the way, some dissatisfaction extraction apparatuses and related apparatuses have been proposed as conventional techniques.

  The information processing device (dissatisfaction extraction device) in Patent Document 1 extracts dissatisfaction regarding disadvantages and disadvantageous events as dissatisfactions that can be used for other services (for example, Q & A sites) from various dissatisfactions. .

  The information processing apparatus of Patent Document 1 acquires a document from a document database, extracts a word representing an improvement operation, extracts a target of the improvement operation from a document acquired from a document set database, and represents a word representing the improvement operation Obtain the co-occurrence frequency of the word representing the improvement action target, obtain the co-occurrence frequency of the word representing the improvement action and the word representing the improvement action target, and deny, neutral, and affirm the word of the improvement action target Classify into any of the evaluation polarities, and output words that indicate the object of negative improvement action as dissatisfied.

  Further, a question posted by the user on the Q & A site is acquired, a question including a word that is dissatisfied is selected from among the questions, a question including a word that is dissatisfied is acquired, and the question is received from the Q & A site. Get the corresponding answer and output it as a recommendation page.

  Specifically, from the explanation sentence such as "This product solves the lack of exercise" described when explaining the effect of the product, the word "elimination" indicating the improvement action and negative action Three words are extracted: “Insufficient exercise” representing the subject and “Resolve me” representing the modality representing profit.

  Next, using the document set, the co-occurrence frequency of the word representing the action and the word representing the object of the action, and the co-occurrence frequency of the word representing the action and the modality representing the profit are obtained, and the two co-occurrence frequencies are obtained. An operation with a large product is extracted as an improvement operation.

  After that, the target of the improved action extracted is identified from the document set, the polarity of the target of the improved action is determined from the co-occurrence frequency of the improved action and the target of the improved action, and finally the target of the improved action whose polarity is negative is determined. Extract as “dissatisfied”.

  A word matching dissatisfaction is extracted from the text written by the user, a question including dissatisfaction is selected from the questions posted on the Q & A site, and an answer associated with the question is presented to the user.

  The comparative evaluation detection apparatus of Patent Literature 2 compares a plurality of objects, and detects a comparative evaluation asserted part composed of a character string expressing a comparative evaluation that describes differences or superiority or inferiority recognized by them.

  The comparative evaluation detection device of Patent Document 2 stores a pattern for extracting a comparative evaluation asserted portion in advance. Next, the text is acquired, divided into character strings of specific units, morphological analysis is performed for each character string, and morphological analysis results are acquired. Then, the morphological analysis result and the extracted pattern are collated. Then, a character string indicating the target, a character string indicating a comparison target when the target is relatively evaluated, a character string indicating an attribute focused on when evaluating the target, and a degree of evaluation regarding the target A combination including either a character string to be indicated or a character string indicating an evaluation related to the target is detected as a comparative evaluation asserted portion.

  The polarity estimation system of Patent Literature 3 estimates the evaluation polarity of reputation information. If this is utilized, it can be estimated which product is evaluated by the magnitude | size of evaluation polarity.

  Based on the reputation information of Patent Document 3 with a known polarity, the evaluation polarity can be estimated and the evaluation polarity is unknown, considering the usage and negation of the expression of good image or bad image in the attribute expression. The polarity can be estimated for an evaluation expression. That is, based on the reputation information accumulated so far, the evaluation polarity can be estimated by taking into account the polarity of the polarity of the object, the attribute expression, and the evaluation expression. Then, a statistical average of the evaluation polarities for the object is obtained as an evaluation value of the object, and an object having a large evaluation value is set as an object evaluated by the user.

JP 2010-204865 A JP 2008-209978 A International Publication No. 2008/077554 Hiroshi Kanayama, Tetsuya Nasukawa, Identification of target of request, The 14th Annual Conference of the Association for Natural Language Processing, Mar. 2008

  The above prior art has the following problems.

  The information processing apparatus of Patent Document 1 extracts dissatisfaction from a sentence describing the effect of a product based on a word representing an improvement operation, and can extract dissatisfaction with respect to the effect aimed at by the company that developed the product. However, it is difficult to extract dissatisfaction useful for improving the evaluation target products from documents written based on the viewpoints of purchasers and users such as blogs and Internet bulletin boards.

  For example, a purchaser or user of a mobile phone complains that the same manufacturer makes it, but mobile phone A can play games but mobile phone B cannot play games. Even if it is written on a bulletin board, etc., it is useful for improving the evaluation target product written by the purchaser or user if the explanation such as “Game play is possible” is not written on the mobile phone. It is not possible to extract unsatisfactory complaints.

  The comparative evaluation and detection device of Patent Document 2 is effective when relatively comparing multiple objects within a certain evaluation axis, such as “The image quality of a liquid crystal television is much better than the image quality of a cathode ray tube.” It was difficult. That is, the comparative evaluation detection device of Patent Document 2 also detects a sentence such as “operability of the mobile phone D is still better than that of the mobile phone C”. The expression “the operability of the mobile phone D is still better than the mobile phone C” includes, in addition, the standard that the user can satisfy both the operability of the mobile phone C and the operability of the mobile phone D. It is assumed that it has not been exceeded. "

  On the other hand, the phrase “the LCD screen of the B company smartphone is beautiful, but the A company smartphone is dirty.” Says, “The user thinks that the LCD screen of the B company smartphone exceeds the standard. However, the liquid crystal screen of the smartphone manufactured by Company A is not considered to exceed the standard. ”Absolute evaluation based on the user's standard is included.

  However, in a relative evaluation such as “the operability of the mobile phone D is still better than that of the mobile phone C”, since the standard that the user can satisfy is not known, the product developer cannot determine how much to improve.

  The polarity estimation system of Patent Document 3 extracts reputation information even when the evaluation object is compared with the comparison object and is not evaluated. For example, even if the user thinks that the mobile phone F is superior to the mobile phone E, if it is not necessary to make a comparative evaluation, the purchaser or the user writes an evaluation such as “the mobile phone E is good”. . I don't know which cell phone is better. When such evaluations are mixed, even if reputation information is collected and a statistical average of evaluation expressions is obtained, the result includes a large error.

  The present invention solves the above problems, and can extract frustrations useful for improving an evaluation target product from documents written by users of the evaluation target product such as blogs and Internet bulletin boards. An object is to provide a dissatisfaction extraction device, method and program.

  The present invention that solves the above-described problem is a sentence including an evaluation object, an evaluation expression related to the evaluation object, a comparison object different from the evaluation object, and an evaluation expression related to the comparison object from a set of sentences. A dissatisfied sentence candidate extraction unit that extracts two consecutive sentences as dissatisfied sentence candidates, a polarity determination unit that determines the degree of polarity of the evaluation expression related to the evaluation object and the evaluation expression related to the comparison object, and the evaluation object And a dissatisfied sentence extraction unit that determines whether or not the dissatisfied sentence candidate is a dissatisfied sentence based on a difference in the degree of polarity between the evaluation expression and the evaluation expression related to the comparison target, and extracts a dissatisfied sentence. It is an unsatisfactory extraction device.

  In the present invention that solves the above-described problem, the dissatisfaction extraction device inputs an evaluation object represented by a character string, and the evaluation object, the evaluation expression related to the evaluation object, and the evaluation object are different from the set of sentences. One sentence including the comparison object and the evaluation expression related to the comparison object or two consecutive sentences are extracted as dissatisfied sentence candidates, and the degree of polarity of the evaluation expression related to the evaluation object and the evaluation expression related to the comparison object is determined. And determining whether or not the dissatisfied sentence candidate is a dissatisfied sentence based on a difference in the degree of polarity between the evaluation expression relating to the evaluation object and the evaluation expression relating to the comparison object, and extracting the dissatisfaction sentence It is a dissatisfaction extraction method.

  The present invention that solves the above-described problems includes an evaluation object input process for inputting an evaluation object represented by a character string, a sentence set, the evaluation object, an evaluation expression related to the evaluation object, and the evaluation object. Dissatisfied sentence candidate extraction processing for extracting different comparison objects, one sentence including the evaluation expression related to the comparison object or two consecutive sentences as dissatisfied sentence candidates, and the evaluation expression related to the evaluation object and the evaluation related to the comparison object Whether or not the dissatisfied sentence candidate is a dissatisfied sentence based on a polarity determination process for determining the degree of polarity of the expression and a difference in the degree of polarity between the evaluation expression relating to the evaluation object and the evaluation expression relating to the comparison object It is a dissatisfaction extraction program characterized by causing a dissatisfaction extraction device to execute a dissatisfaction extraction process for determining and extracting dissatisfaction sentences.

  ADVANTAGE OF THE INVENTION According to this invention, the dissatisfaction useful for improvement of evaluation object product can be extracted from the document written by the user of evaluation object product like a blog or an internet bulletin board. Product developers can immediately feed back the extracted complaints to product development.

Functional block diagram of the dissatisfaction extraction device (first embodiment) Polar dictionary structure Flow chart showing treatment details of dissatisfaction extraction device Functional block diagram of dissatisfaction extraction device (modification) Flow chart showing treatment details of dissatisfaction extraction device (second embodiment) Flow chart showing treatment details of dissatisfaction extraction device Structure of upper and lower relation dictionary Functional block diagram of dissatisfaction extraction device (modification)

<First Embodiment>
~Constitution~
The configuration of the first embodiment will be described. FIG. 1 is a functional block diagram of the dissatisfaction extraction device according to the first embodiment. The dissatisfaction extracting apparatus inputs a sentence set acquisition unit 101 that acquires and stores a set of sentences, an evaluation target input unit 102 that inputs an evaluation target as a product to be improved, and a comparison target corresponding to the evaluation target From the comparison target input unit 103 and the set of sentences, one evaluation sentence or two consecutive sentences including the evaluation expression related to the evaluation object and the evaluation object and the evaluation expression related to the comparison object and the comparison object are extracted as dissatisfied sentence candidates. The dissatisfied sentence candidate extraction unit 104, the polarity determination unit 105 that determines the degree of polarity of the evaluation expression related to the evaluation target and the evaluation expression related to the comparison target, and the polarity of the evaluation expression related to the evaluation target and the evaluation expression related to the comparison target Based on the degree difference, it is determined whether or not the dissatisfied sentence candidate is a dissatisfied sentence, and includes a dissatisfied sentence extracting unit 106 that extracts a dissatisfied sentence, and a dissatisfied storage unit 107 that stores the extracted dissatisfied sentence. That.

  The sentence set acquisition unit 101 acquires and stores a set of sentences from a blog, an internet bulletin board, a word-of-mouth site, etc. written by the purchaser or user of the target product.

  The evaluation target input unit 102 inputs a character string input with a keyboard or the like as an evaluation target. In this embodiment, “A company-made smartphone” is an evaluation target as an example.

  The comparison target input unit 103 inputs a character string input with a keyboard or the like as a comparison target. In the present embodiment, for example, “B company smartphone” or “C company smartphone” is used as a comparison target.

  The dissatisfied sentence candidate extraction unit 104 searches the evaluation set related to the evaluation target and the evaluation expression related to the comparison target from the set of sentences stored in the sentence set acquisition unit 101. The evaluation expression is determined with reference to the evaluation expression dictionary 108. One sentence or two consecutive sentences including the evaluation expression related to the evaluation object and the evaluation object and the evaluation expression related to the comparison object and the comparison object are extracted as dissatisfied sentence candidates.

For example, a set of sentences including the following sentences (a) to (j) is assumed. An evaluation expression related to the evaluation target and an evaluation expression related to the comparison target are indicated by [].
(A) In the case of the smartphone manufactured by B company, the electronic money function [can be used], but in the smartphone manufactured by company A, the electronic money function cannot be used.
(B) Although the FM transmission function [can be used] like a smartphone manufactured by B company, the FM transmission function [cannot be used] by a smartphone manufactured by A company.
(C) A smartphone made by company A is [dirty], but a smartphone made by company B has a clean screen.
(D) The screen of the smartphone manufactured by Company A is [good]. The screen of the B company smartphone [cannot be complained].
(E) The screen of the C company smartphone is [good]. The screen of the smartphone manufactured by Company A [cannot be complained].
(F) The screen of the smartphone manufactured by Company A is also good. The screen of the C company smartphone is also good.
(G) Although the screen of the C company smartphone is [not good], the screen of the company A smartphone is [too bad].
(H) The smartphone manufactured by Company A normally [cannot be charged] on the second day, and the red lamp blinked.
(I) I would like to be able to automatically refresh the screen with a company A smartphone.
(J) Although mobile phones are foldable, they cannot be used with smartphones.

  That is, the dissatisfied sentence candidate extracting unit 104 extracts each sentence of (a) to (g) as a dissatisfied sentence candidate.

  In (h) and (i), there is no evaluation expression related to the comparison object and the comparison object. In (j), there is no evaluation expression related to the evaluation object and the evaluation object, and therefore, it is excluded from the candidates.

  The polarity determination unit 105 determines the degree of polarity of the evaluation expression related to the evaluation object and the evaluation expression related to the comparison object. The evaluation polarity is a characteristic of whether the evaluation expression is a positive expression or a negative expression. The polarity of the evaluation expression is determined with reference to the polarity dictionary 109.

FIG. 2 is a diagram illustrating the structure of the polarity dictionary 109. “[Sa variation] can be” or the like represents a pattern of “use (sa variation)” + “can” such as “can be used”. It is assumed that the polarity is a real number from −1 to 1, and the closer to 1, the more positive, and the closer to −1, the negative. Polarity = 0 indicates neutral evaluation. That is, -1 ≦ polarity <0 indicates negative evaluation, and 0 <polarity ≦ 1 indicates positive evaluation. The polarity of the evaluation expression included in the candidates (a) to (g) is determined.
(A) [can be used] = 0.6, [cannot] = − 1.0
(B) [Available] = 0.6, [Unusable] = − 1.0
(C) [Dirty] = − 0.9, [Beautiful] = 0.9
(D) [Good] = 0.5, [No complaining] = 1.0
(E) [Good] = 0.5, [No complaining] = 1.0
(F) [Good] = 0.5, [Good] = 0.5
(G) [not good] = − 0.3, [too severe] = − 1.0

When the evaluation expression related to the comparison target is a positive evaluation expression or a neutral evaluation expression (Pc ≧ 0), the dissatisfaction sentence extraction unit 106 evaluates the value Pt indicating the degree of polarity of the evaluation expression related to the evaluation target and the comparison target A difference (Pc−Pt) from the value Pc of the degree of polarity of the expression is obtained.
(A) Pc-Pt = 0.6-(-1.0) = 1.6
(B) Pc-Pt = 0.6-(-1.0) = 1.6
(C) Pc−Pt = 0.9 − (− 0.9) = 1.8
(D) Pc-Pt = 1.0-0.5 = 0.5
(E) Pc-Pt = 0.5-1.0 = -0.5
(F) Pc-Pt = 0.5-0.5 = 0.0

  In addition, in (g), since the evaluation expression [not good] concerning the comparison target “smart phone made by company C” is negative evaluation (= −0.3 <0.0), it is excluded. This is because even if the screen of the smartphone manufactured by company A is improved to have the same quality as the screen of the smartphone manufactured by company C, the user feels [not good enough] and is not satisfied.

  The polarity difference (Pc−Pt) is an index representing the user's expectation for improvement. If this index value increases, the effect of improvement increases. The threshold value is a value that can show a clear superiority or inferiority.

  The dissatisfied sentence extraction unit 106 determines that the sentences (a) to (d) in which the polarity difference (Pc−Pt) is equal to or greater than a predetermined threshold (for example, 0.4) are dissatisfied sentences.

  In (e), the polarity difference (Pc−Pt) is a negative number. This means that the evaluation target is superior to the comparison target, and there is no improvement. Further, in (f), the polarity difference (Pc−Pt) is less than the threshold value, there is no superiority or inferiority difference between the evaluation object and the comparison object, and there is no improvement point.

  The dissatisfaction storage unit 107 stores dissatisfaction sentences (a), (b), (c), and (d).

  The product developer accesses the dissatisfaction storage unit 107 and refers to the dissatisfaction, thereby providing immediate feedback on product improvement.

~ Operation ~
FIG. 3 is a flowchart showing the treatment content of the dissatisfaction extraction device. The operation of the dissatisfaction extraction device will be described using a flowchart.

  First, a set of sentences is acquired and stored from a blog, an internet bulletin board, a word-of-mouth website, etc. written by the purchaser or user of the target product (step 1). For example, a set of sentences including the sentences (a) to (j) above is acquired.

  On the other hand, a character string input with a keyboard or the like is input as an evaluation target (step 2). For example, “Smartphone manufactured by Company A” is input as an evaluation target. In addition, a character string input with a keyboard or the like is input as a comparison target (step 3). For example, “B company smartphone” and “C company smartphone” are input as comparison targets.

  From a set of sentences, an evaluation object, an evaluation expression related to the evaluation object, a comparison object different from the evaluation object, and one sentence or two consecutive sentences containing the evaluation expression related to the comparison object (Step 4). For example, each sentence of (a) to (g) is extracted as a dissatisfied sentence candidate.

  Next, the degree of polarity of the evaluation expression related to the evaluation object and the evaluation expression related to the comparison object is determined (step 5). Refer to the polarity dictionary 109 for the polarity of the evaluation expression.

  The evaluation expressions related to the comparison target “B company smartphone” and “C company smartphone” are positive evaluation expressions or neutral evaluation expressions, and indicate the degree of polarity of the evaluation expression related to the evaluation object “A company smartphone”. It is determined that each sentence of the dissatisfied sentence candidates (a) to (d) whose difference between the value and the value of the degree of polarity of the evaluation expression related to the comparison target is equal to or greater than a predetermined threshold is a dissatisfied sentence, Extract (step 6) and store (step 7).

  These processes are executed by a program of the dissatisfaction extraction device.

~effect~
With the dissatisfaction extraction device of the present embodiment, it is possible to extract dissatisfied sentences in which comparison targets with excellent evaluations and evaluation targets with inferior evaluations are described in contrast expressions. For example, in the dissatisfied sentences such as the sentences (a) to (c) above, “the user thinks that the comparison target exceeds the standard, but the evaluation target does not exceed the standard. The absolute evaluation of "I think" is implied.

  That is, the product developer can make improvements that exceed the standards satisfied by the user by adding the functions evaluated in the comparison target to the evaluation target. For example, an electronic money function, an FM transmission function, and a music file playback function may be added to a smartphone manufactured by company A to improve the image quality of the screen. The function has already been realized in the smartphone manufactured by Company B, which is a comparison target, and there is almost no technical problem in adding the function to the evaluation target.

  Moreover, the dissatisfaction extraction apparatus of this embodiment can also extract the dissatisfaction sentence in which the potential dissatisfaction is described. For example, in the dissatisfaction sentence like each sentence of (d) above, the user does not feel direct dissatisfaction with the smartphone manufactured by company A which is the object of evaluation and exceeds the standard, but B which is the object of comparison. We feel that company-made smartphones are superior, and users will be more satisfied if the screen of smartphones made by company A is improved to the same quality as smartphones made by company B. The function has already been realized in the smartphone manufactured by Company B, which is a comparison target, and there is almost no technical problem in adding the function to the evaluation target.

  In this way, the product developer can immediately feed back user dissatisfaction to product improvement and new product development.

~ Modification ~
In this embodiment, the dissatisfied sentence candidate extraction unit 104 searches the evaluation expression related to the evaluation target and the evaluation expression related to the comparison target from the sentence set stored in the sentence set acquisition unit 101. The evaluation expression is determined with reference to the evaluation expression dictionary 108. One sentence or two consecutive sentences including the evaluation expression related to the evaluation object and the evaluation object and the evaluation expression related to the comparison object and the comparison object are extracted as dissatisfied sentence candidates.

  As a modified example, an inverse expression may be detected between the evaluation object and the comparison object. FIG. 4 is a functional block diagram of a dissatisfaction extraction device according to a modification.

  The dissatisfied sentence candidate extraction unit 104 detects a reverse expression from the evaluation target and the comparison target from the sentence set stored in the sentence set acquisition unit 101. Whether or not the reverse connection expression is used is determined by referring to the reverse connection expression dictionary 110. One sentence or two consecutive sentences including the evaluation expression related to the evaluation object and the evaluation object, the evaluation expression related to the comparison object and the comparison object, and the reverse connection expression are extracted as dissatisfied sentence candidates.

In each of the sentences (a) to (j) above, the reverse connection expression is indicated by <<>>.
(A) In the case of a smartphone manufactured by B company, the electronic money function [can be used] << but >>, and in the smartphone manufactured by company A, the electronic money function cannot be used.
(B) The FM transmission function [can be used] like the smartphone manufactured by B company, but the FM transmission function [cannot be used] by the smartphone manufactured by company A.
(C) A smartphone made by company A has [dirty] << ga >>, and a smartphone made by company B has a beautiful screen.
(D) The screen of the smartphone manufactured by Company A is [good]. The screen of the B company smartphone [cannot be complained].
(E) The screen of the C company smartphone is [good]. The screen of the smartphone manufactured by Company A [cannot be complained].
(F) The screen of the smartphone manufactured by Company A is also good. The screen of the C company smartphone is also good.
(G) The screen of the smartphone manufactured by company C is [not good] << but >>, and the screen of the smartphone manufactured by company A is [too bad].
(H) The smartphone manufactured by Company A normally [cannot be charged] on the second day, and the red lamp blinked.
(I) I would like to be able to automatically refresh the screen with a company A smartphone.
(J) Mobile phones are foldable, but not with smartphones.

  That is, the dissatisfied sentence candidate extraction unit 104 extracts the sentences (a) to (c) and (g) as dissatisfied sentence candidates. Other configurations and operations are the same as in this embodiment.

  Thereby, a dissatisfied sentence candidate can be extracted accurately. As a result, dissatisfied sentences can be extracted with high accuracy.

  Another modification will be described.

  In the present embodiment, in order to exclude the part (g) described above, the dissatisfied sentence extraction unit 106 operates only when the evaluation expression related to the comparison target is a positive evaluation expression or a neutral evaluation expression (Pc ≧ 0). To do. This is because even if the screen of the smartphone manufactured by company A is improved to have the same quality as the screen of the smartphone manufactured by company C, the user feels [not good enough] and is not satisfied. In the present embodiment, such dissatisfaction is not useful, but may be useful in some cases. In that case, the dissatisfied sentence extraction unit 106 operates by canceling the condition of (Pc ≧ 0).

  In this embodiment, in order to extract a clear superiority / inferiority difference, the dissatisfied sentence extraction unit 106 determines that a dissatisfied sentence candidate having a polarity difference (Pc−Pt) equal to or greater than a threshold is a dissatisfied sentence. However, if the superiority difference can be extracted accurately by making the polarity dictionary 109 more accurate, it may be determined that a dissatisfied sentence candidate with a polarity difference (Pc−Pt)> 0 is a dissatisfied sentence.

Second Embodiment
FIG. 5 is a functional block diagram of the dissatisfaction extraction device according to the second embodiment, and FIG. 6 is a flowchart showing the treatment content of the dissatisfaction extraction device according to the second embodiment.

  In the first embodiment, the comparison target input unit 103 inputs a character string input with a keyboard or the like as a comparison target (step 3), but in the second embodiment, the comparison target specifying unit 111 has an upper-lower relationship. The comparison target corresponding to the evaluation target is specified with reference to the dictionary 112 (step 8).

  FIG. 7 is a diagram showing the structure of the upper / lower relationship dictionary 112. There are “smartphone”, “mobile phone”, and “landline phone” as subordinate terms of “telephone”. As a subordinate term of “smartphone”, “smartphone manufactured by company A”, “smartphone manufactured by company B”, “smartphone manufactured by company C”, and the like exist. Conversely, “smartphone” exists as a broad term for “smart phone made by company A”, “smart phone made by company B”, and “smart phone made by company C”, and “phone” exists as a broad term for “smartphone”. You can also.

  In order to build the upper-lower relationship, the upper-lower relationship extraction tool distributed in the advanced language information fusion forum can be used.

  “Smartphone manufactured by Company A”, “Smartphone manufactured by Company B”, and “Smartphone manufactured by Company C” are the same terms as the broader term “smartphone”.

  The comparison target specifying unit 111 specifies a broader word to be evaluated, and specifies a narrower word of the broader word as a comparison target. For example, “smart phone” is specified as a broad term of the evaluation target “smart phone made by company A”, and “smart phone made by company B” and “smart phone made by company C”, which are subordinate terms of “smart phone”, are identified as comparison targets. Then, it is confirmed whether or not a word representing the comparison object appears within N morphemes from the word representing the evaluation object, and the comparison object is extracted. Here, N represents a co-occurring range, and N is an arbitrary value (for example, N = 100 is set).

  The comparison target specifying unit 111 can save time and effort for inputting a comparison target. Furthermore, since the comparison target specifying unit 111 requests that the comparison target appears in the vicinity of the evaluation target, it is possible to extract a dissatisfied sentence candidate with higher accuracy.

~ Modification ~
FIG. 8 is a functional block diagram of a dissatisfaction extraction device according to a modification of the second embodiment.

  In the modification, the sentence set acquisition unit 101 in the second embodiment is a sentence set acquisition unit 113, and there is no evaluation target input unit 102.

  The sentence set acquisition unit 113 acquires and stores a set of sentences from a blog, an Internet bulletin board, a word-of-mouth site, and the like related to the evaluation target “Company A” written by the purchaser or user of the target product.

  The comparison target specifying unit 111 specifies the comparison target corresponding to the evaluation target with reference to the upper and lower relationship dictionary 112. For example, “B company smartphone” or “C company smartphone” is the comparison target.

  Other configurations and operations are the same as those in the second embodiment.

  Blogs, Internet bulletin boards, and word-of-mouth sites related to the evaluation target “Smartphone manufactured by Company A” are premised on topics related to the evaluation target, and the evaluation target is often omitted. In the modified example, the sentence set acquisition unit 116 estimates that the description of the evaluation target is omitted by acquiring a set of sentences from a blog, an internet bulletin board, a word-of-mouth website, or the like regarding the evaluation target “Company A smartphone”. can do.

  Moreover, the trouble of inputting the evaluation object can be saved by the sentence set acquisition unit 113.

<Appendix>
A part or all of the above embodiment may be described as an appendix as follows, but is not limited to the following.

  The present invention provides an evaluation object, an evaluation expression related to the evaluation object, a comparison object different from the evaluation object, and one sentence or two consecutive sentences including the evaluation expression related to the comparison object from a set of sentences. A dissatisfied sentence candidate extracting unit that extracts a dissatisfied sentence candidate, a polarity determining unit that determines the degree of polarity of the evaluation expression related to the evaluation object and the evaluation expression related to the comparison object, and the evaluation expression and the comparison object related to the evaluation object And a dissatisfied sentence extraction unit for determining whether or not the dissatisfied sentence candidate is a dissatisfied sentence based on a difference in the degree of polarity of the evaluation expression according to It is.

  With the dissatisfaction extraction device configured as described above, it is possible to extract a dissatisfied sentence in which the evaluation object to be evaluated is superior and the evaluation object to be evaluated to be inferior are described in contrast. The dissatisfaction statement implicitly states an absolute evaluation that the user thinks that the comparison target exceeds the standard, but the evaluation target does not exceed the standard. Yes.

  That is, the product developer can make improvements that exceed the standards satisfied by the user by adding the functions evaluated in the comparison target to the evaluation target. The function has already been realized in the comparison target that is the comparison target, and there is almost no technical problem in adding the function to the evaluation target.

  Moreover, the dissatisfaction extraction apparatus of this invention can also extract the dissatisfaction sentence in which the potential dissatisfaction is described. That is, the user does not feel direct dissatisfaction with the evaluation object and exceeds the standard, but if the user feels that the comparison object is superior, the user becomes a potential dissatisfaction. If the product developer adds various functions evaluated in the comparison object to the evaluation object, the user is more satisfied. The function has already been realized in the comparison target that is the comparison target, and there is almost no technical problem in adding the function to the evaluation target.

  In this way, the product developer can immediately feed back user dissatisfaction to product improvement and new product development.

  In the dissatisfaction extraction device of the present invention, more preferably, the dissatisfaction sentence extraction unit calculates a difference between a value indicating the degree of polarity of the evaluation expression related to the evaluation object and a value of the degree of polarity of the evaluation expression related to the comparison object. If the difference between the polarity values is equal to or greater than a predetermined threshold, it is determined that the dissatisfied sentence candidate is a dissatisfied sentence.

  Thereby, the dissatisfied sentence with a clear superiority difference between a comparison object and an evaluation object can be extracted more accurately.

  In the dissatisfied extraction device of the present invention, more preferably, the dissatisfied sentence extracting unit further determines that the dissatisfied sentence candidate is a dissatisfied sentence if the evaluation expression related to the comparison target is a positive evaluation expression or a neutral evaluation expression. judge.

  If the comparison target does not exceed the standard, the user is not satisfied even if improvements are made by adding various functions evaluated by the comparison target to the evaluation target. Such dissatisfaction that is not useful for improving the evaluation target product can be excluded.

  In the dissatisfaction extraction device of the present invention, more preferably, the dissatisfied sentence candidate extraction unit further extracts a sentence including a reverse tangent expression between the evaluation object and the comparison object as a dissatisfied sentence candidate.

  In the dissatisfaction extraction device according to the present invention, it is more preferable that the dissatisfaction extraction device further includes a comparison target specifying unit that specifies the broader word to be evaluated and specifies the lower word of the broader word as the comparison target.

  In the present invention, the dissatisfaction extraction device inputs an evaluation object represented by a character string, and from a set of sentences, the evaluation object, an evaluation expression related to the evaluation object, a comparison object different from the evaluation object, One sentence or two consecutive sentences including the evaluation expression related to the comparison target are extracted as dissatisfied sentence candidates, the degree of polarity of the evaluation expression related to the evaluation target and the evaluation expression related to the comparison target is determined, and the evaluation target A method for extracting dissatisfaction is characterized by determining whether or not the dissatisfied sentence candidate is a dissatisfied sentence based on a difference in the degree of polarity between the evaluation expression relating to the evaluation expression and the evaluation expression relating to the comparison target, and extracting the dissatisfaction sentence It is.

  In the dissatisfaction extraction method of the present invention, more preferably, in the step of extracting the dissatisfaction sentence, a value indicating a degree of polarity of the evaluation expression related to the evaluation object and a value of a degree of polarity of the evaluation expression related to the comparison object A difference is obtained, and if the difference between the polarity values is equal to or greater than a predetermined threshold, it is determined that the dissatisfied sentence candidate is a dissatisfied sentence.

  In the dissatisfaction extraction method of the present invention, more preferably, in the step of extracting the dissatisfied sentence, if the evaluation expression related to the comparison target is a positive evaluation expression or a neutral evaluation expression, the dissatisfied sentence candidate is a dissatisfied sentence. Judge that there is.

  In the dissatisfaction extraction method of the present invention, more preferably, in the step of extracting the dissatisfied sentence candidate, a sentence including an inverse expression between the evaluation object and the comparison object is further extracted as a dissatisfied sentence candidate.

  In the dissatisfaction extraction method of the present invention, more preferably, the broader word to be evaluated is identified, and the narrower word of the broader word is identified as the comparison target.

  The present invention relates to an evaluation object input process for inputting an evaluation object represented by a character string, a sentence set, the evaluation object, an evaluation expression related to the evaluation object, and a comparison object different from the evaluation object, Dissatisfied sentence candidate extraction process for extracting one sentence or two consecutive sentences including the evaluation expression related to the comparison object as a dissatisfied sentence candidate, and the degree of polarity of the evaluation expression related to the evaluation object and the evaluation expression related to the comparison object And determining whether or not the dissatisfied sentence candidate is a dissatisfied sentence based on a difference in polarity between the polarity determination process for determining and the evaluation expression relating to the evaluation object and the evaluation expression relating to the comparison object, and dissatisfaction sentence This is a dissatisfaction extraction program that causes a dissatisfaction extraction device to execute a dissatisfaction sentence extraction process for extracting.

  In the dissatisfaction extraction program of the present invention, more preferably, in the dissatisfaction sentence extraction process, a difference between a value indicating the degree of polarity of the evaluation expression related to the evaluation object and a value of the degree of polarity of the evaluation expression related to the comparison object is calculated. If the difference between the polarity values is equal to or greater than a predetermined threshold, it is determined that the dissatisfied sentence candidate is a dissatisfied sentence.

  In the dissatisfaction extraction program of the present invention, more preferably, in the dissatisfied sentence extraction process, if the evaluation expression related to the comparison target is a positive evaluation expression or a neutral evaluation expression, the dissatisfied sentence candidate is a dissatisfied sentence. judge.

  In the dissatisfaction extraction program of the present invention, more preferably, in the dissatisfied sentence candidate extraction process, a sentence including a reverse tangent expression between the evaluation object and the comparison object is further extracted as a dissatisfied sentence candidate.

In the dissatisfaction extraction program according to the present invention, more preferably, the dissatisfaction extracting apparatus further executes a comparison object specifying process for specifying the broader word to be evaluated and specifying the low-order word of the broader word as the comparison object.
This application claims the priority on the basis of Japanese application Japanese Patent Application No. 2012-041395 for which it applied on February 28, 2012, and takes in those the indications of all here.

101 sentence set acquisition unit 102 evaluation target input unit 103 comparison target input unit 104 dissatisfied sentence candidate extraction unit 105 polarity determination unit 106 dissatisfaction sentence extraction unit 107 dissatisfaction storage unit 108 evaluation expression dictionary 109 polarity dictionary 110 reverse connection expression dictionary (modified example)
111 Comparison Target Identification Unit (Second Embodiment)
112 Upper / Lower Relation Dictionary (Second Embodiment)
113 sentence set acquisition part (modification)

Claims (7)

From the set of statements, the evaluation expression of the evaluation objects and the evaluation objects, according to one sentence or evaluation objects and the evaluation objects, and an evaluation expression according to different comparison and the comparison subject with the evaluation A dissatisfied sentence candidate extraction unit that extracts, as dissatisfied sentence candidates , two consecutive sentences including a sentence including an evaluation expression and a comparison object different from the evaluation object and a sentence including the evaluation expression related to the comparison object ;
A polarity determination unit that determines the degree of polarity of the evaluation expression related to the evaluation object and the evaluation expression related to the comparison object;
A dissatisfied sentence extraction unit for determining whether or not the dissatisfied sentence candidate is a dissatisfied sentence based on a difference in the degree of polarity between the evaluation expression relating to the evaluation object and the evaluation expression relating to the comparison object; A dissatisfaction extraction device characterized by comprising: The dissatisfaction extraction device according to claim 1,
The dissatisfied sentence extraction unit
Find the difference between the value indicating the degree of polarity of the evaluation expression related to the evaluation object and the value of the degree of polarity of the evaluation expression related to the comparison object,
The dissatisfaction extraction device, wherein the dissatisfied sentence candidate is determined to be a dissatisfied sentence if the difference between the polarity values is equal to or greater than a predetermined threshold. In the dissatisfaction extraction apparatus in any one of Claims 1-2,
The dissatisfied sentence extraction unit
Furthermore, if the evaluation expression which concerns on the said comparison object is a positive evaluation expression or a neutral evaluation expression, it determines with the said dissatisfied sentence candidate being a dissatisfied sentence. The dissatisfaction extraction apparatus characterized by the above-mentioned. In the dissatisfaction extraction apparatus in any one of Claims 1-3,
The dissatisfied sentence candidate extraction unit
Furthermore, the dissatisfaction extraction apparatus characterized by extracting the sentence containing reverse connection expression between the said evaluation object and the said comparison object as a dissatisfied sentence candidate. In the dissatisfaction extraction apparatus in any one of Claims 1-4,
A dissatisfaction extraction device, further comprising: a comparison target specifying unit that specifies a broader word to be evaluated and specifies a lower word of the broader word as the comparison target. A dissatisfaction extraction method using a dissatisfaction extraction device comprising an evaluation target input part, a dissatisfied sentence candidate extraction part, a polarity determination part, and a dissatisfaction sentence extraction part,
The evaluation target input part inputs the evaluation target represented by the character string,
Dissatisfaction sentence candidate extraction unit, from a set of statements, the evaluation expression of the evaluation objects and the evaluation target, or one sentence, and a evaluation expressions according to different comparison and the comparison subject with the evaluation, evaluation a sentence containing the evaluation expression of the target and the evaluation objects, the two sentences continuous comprising a sentence including the evaluation expressions of the different comparison and the comparison subject with the evaluation target, and extracted as dissatisfaction sentence candidates,
The polarity determination unit determines the degree of polarity of the evaluation expression related to the evaluation object and the evaluation expression related to the comparison object,
The dissatisfied sentence extraction unit determines whether or not the dissatisfied sentence candidate is a dissatisfied sentence based on the difference in polarity between the evaluation expression related to the evaluation object and the evaluation expression related to the comparison object, and extracts the dissatisfied sentence Dissatisfaction extraction method characterized by A dissatisfaction extraction program that causes a dissatisfaction extraction apparatus including an evaluation target input unit, a dissatisfied sentence candidate extraction unit, a polarity determination unit, and a dissatisfaction sentence extraction unit to perform information processing,
Evaluation input unit, and the evaluation object input process of inputting evaluated represented by the string,
Dissatisfaction sentence candidate extraction unit, from a set of statements, the evaluation expression of the evaluation objects and the evaluation target, or one sentence, and a evaluation expressions according to different comparison and the comparison subject with the evaluation, evaluation complaints of extracting a sentence including the evaluation expression of the target and the evaluation objects, the two sentences continuous comprising a sentence including the evaluation expressions of the different comparison and the comparison subject with the evaluation, as dissatisfaction sentence candidates Sentence candidate extraction processing;
A polarity determination unit for determining the degree of polarity of the evaluation expression related to the evaluation object and the evaluation expression related to the comparison object;
The dissatisfied sentence extraction unit determines whether or not the dissatisfied sentence candidate is a dissatisfied sentence based on the difference in polarity between the evaluation expression related to the evaluation object and the evaluation expression related to the comparison object, and extracts the dissatisfied sentence A dissatisfaction extraction program for causing a dissatisfaction extraction device to execute dissatisfaction sentence extraction processing.

Images