JPWO2015118683A1 - Opinion collection device and system, and opinion collection method - Google Patents

Abstract

When collecting opinions on a topic, set multiple values related to the topic so that the commenter of the opinion can understand the background of the way of thinking about values. Input a topic and values related to the topic, search for an opinion on the topic, search for past opinions of the utterances of the opinion, calculate importance on the values of past utterances, and display the searched opinions. Based on the degree of importance of multiple values, the subject is mapped to the display unit, and individual opinions on the topic are displayed in association with the subject. While referring to each opinion. The importance of values is determined by finding the places where expressions with values and semantic similarity appear in the past utterances of the utterance subject, and the synergistic value of the semantic similarity and the context factor of each expression at each occurrence is passed to the past utterances. To calculate.

Description

The present invention relates to an opinion collection device and system, and an opinion collection method.

When analyzing or making a decision on a topic, there is an effort to collect the various opinions from the respective pros and cons of the topic, and to make an optimal decision by comparing the lengths of each. Generally done. Technologies have been developed to support these efforts.
Patent Document 1 describes a technique for mapping characteristic words of pros and cons according to pros and cons specificity and topic specificity.

JP 2007-241901 A

When analyzing or making a decision on a topic, there is an effort to collect the various opinions from the respective pros and cons of the topic, and to make an optimal decision by comparing the lengths of each. Generally done. At that time, if it is possible to know the background of how much importance is given to a plurality of values related to the topic in general by the subject of each opinion, a more accurate analysis can be performed.
For example, in the case of a topic that asks the right or wrong of animal experiments for the development of new medicines, from the standpoint of emphasizing medical progress, animal experiments are often considered to be necessary, and animal welfare and animal bioethics are also considered. From the point of emphasis, it is often thought that it should be prohibited. Therefore, in this case, it is considered that there is a source of the topic where the two values of medical progress and animal bioethics are not compatible.
However, if we leave the topic of animal experimentation, it is common sense that both the advancement of medicine and the protection of animal life are usually important. In this way, what many people consider to be important in common sense when they are questioned about good and evil independently from other values is called values (or values). .

Suppose now that Mr. A, Mr. B, Mr. C, and Mr. D stated that animal experiments are necessary. In this case, if Mr. A, Mr. B, Mr. C, and Mr. D are independent of the subject of animal experiments, the values of "medical progress" and the values of "animal bioethics" are considered as independent values. If it can be quantitatively estimated from past remarks etc. whether there is importance or not, there is an effect that even the same opinion that “animal experiment is necessary” will be seen differently.
For example, if Mr. A has a high degree of emphasis on medical progress, but is less interested in animal bioethics, then he would be construed as a “Background” opinion. If you are a person who has a strong interest in ethics, this opinion can be viewed as an “unexpectedly different opinion”. It is also possible that Mr. C may have just said “slightly” if the importance of both medical progress and animal ethics is low from past remarks. On the contrary, if Mr. D has a high degree of emphasis on both values, this opinion may be considered as “a thoughtful opinion based on his own weight”.
The ability to quantitatively estimate the degree of importance regarding values in this way has great value in opinion analysis, but has not been realized so far. One of the problems to be solved by the present invention is to provide means for realizing this.
Here, A to D are individuals, but in reality, there are cases where the organization is an organization such as a website or a medium such as a magazine. .

In view of the above points, the present invention calculates and displays or outputs a plurality of opinions relating to a given topic by calculating the degree of importance (background, etc.) of each speaker's way of thinking regarding a plurality of values related to the topic. With the goal.

According to the first solution of the present invention,
An opinion collecting device,
A plurality of document data including the document content in advance, a plurality of opinion data including the document content and the speaking subject, a storage unit holding importance data for each speaking subject and each value,
An arithmetic unit;
With

The computing unit is
Receiving a topic entered by the terminal that defines what documents to collect and one or more values that are considered to have an impact when determining the pros and cons of the topic;
Search the document data for the document content related to the received topic,
Finding a set of speaking subjects of the retrieved document content, storing a plurality of opinion data including the document content and the speaking subject in the storage unit,
For each speaking subject included in the opinion data, calculate the importance for each of the values,
Create importance level data for each subject and value from the calculated importance level, store it in the storage unit,
An opinion collecting apparatus is provided, characterized in that the importance data classified by each speaking subject and each value is displayed on a display unit or output on an output unit.

According to the second solution of the present invention,
An opinion collection system,
A terminal,
An opinion collection device connected to the terminal via a communication network;

The opinion collection device includes:
A plurality of document data including the document content in advance, a plurality of opinion data including the document content and the speaking subject, a storage unit holding importance data for each speaking subject and each value,
An arithmetic unit;
Have

The computing unit is
Receiving a topic entered by the terminal that defines what documents to collect and one or more values that are considered to have an impact when determining the pros and cons of the topic;
Search the document data for the document content related to the received topic,
Finding a set of speaking subjects of the retrieved document content, storing a plurality of opinion data including the document content and the speaking subject in the storage unit,
For each speaking subject included in the opinion data, calculate the importance for each of the values,
Create importance level data for each subject and value from the calculated importance level, store it in the storage unit,
There is provided an opinion collection system characterized in that the importance data for each subject and each value is displayed on a display unit or output on an output unit.

According to the third solution of the present invention,
An opinion collection method in an opinion collection device,
The opinion collection device includes:
A plurality of document data including the document content in advance, a plurality of opinion data including the document content and the speaking subject, a storage unit holding importance data for each speaking subject and each value,
An arithmetic unit;
With

The computing unit is
Receiving a topic entered by the terminal that defines what documents to collect and one or more values that are considered to have an impact when determining the pros and cons of the topic;
Search the document data for the document content related to the received topic,
Finding a set of speaking subjects of the retrieved document content, storing a plurality of opinion data including the document content and the speaking subject in the storage unit,
For each speaking subject included in the opinion data, calculate the importance for each of the values,
Create importance level data for each subject and value from the calculated importance level, store it in the storage unit,
An opinion collecting method is provided, characterized in that the importance data classified by the subject and the value is displayed on a display unit or output on an output unit.

According to the present invention, it is possible to calculate and display or output a plurality of opinions relating to a given topic by calculating the degree of importance (background etc.) of each speaker's way of thinking regarding a plurality of values related to the topic.

It is a block diagram of the opinion collection system which concerns on one Example of this invention. It is a figure which shows the detail of the example of an opinion browsing analysis support screen 251 displayed on the display part of FIG. 1, and the detail of the opinion browsing analysis support work area 2212. It is a figure which shows the structure of the value data stored in the value storage area of FIG. It is a figure which shows the structure of the opinion data stored in the opinion data storage area of FIG. It is a figure which shows the structure of the speech subject data stored in the speech subject data storage area of FIG. It is a figure which shows the structure of the speech subject x importance by value storage stored in the message subject x sense of importance storage area of FIG. It is a block diagram of the opinion collection work area 1215 of an opinion collection device. It is a block diagram of the rule and data collection 1216 for opinion collection of an opinion collection device. It is a figure which shows the sequence in which the opinion browsing analysis support management part and opinion collection management part of FIG. 1 operate | move via a communication network. It is a figure explaining the procedure in which the opinion collection means 1211 collects the opinion regarding a topic from a given topic in step F4 of FIG. It is a figure which shows an example of approval / disagreement expression data used in the case of opinion collection. It is a figure which shows an example of reason and evidence expression data used in the case of opinion collection. It is a figure which shows an example of the negative expression data used in the case of opinion collection. It is a figure which shows an example of the utterance expression data used in the case of opinion collection. A procedure in which the value importance degree calculation unit 1213 collects the past remarks of a given subject and the given values and calculates the importance on the values from the collected remarks will be described. FIG. It is a figure which shows an example of the semantic similarity phrase pair data used at the time of importance calculation. It is a figure which shows an example of promotion / suppression word / phrase data used at the time of importance calculation. It is the figure which showed the content of the sentence structure storage area obtained as a result of analyzing the example sentence stored in the sentence storage area at the time of importance calculation, and its syntax structure. It is the figure which showed the content of the context factor calculation work area at the time of calculating importance regarding the same example sentence as FIG. It is a figure which shows an example of the main part factor data used at the time of context factor calculation. It is a figure which shows an example of auxiliary | assistant part factor data used at the time of context factor calculation. It is a figure which shows an example of modification part factor data used at the time of context factor calculation. It is a screen example of the opinion browsing analysis support screen 251 provided with the age division setting part. This is a screen example when opinions about products and services are collected based on price and performance as values. It is an example of a screen when opinions other than the concerned country are collected as values based on the relations with the concerned countries (A and B) regarding diplomatic disputes. It is a figure which shows an example of the document data 131 of FIG. It is a figure which shows an example of the search index data 132 of FIG. It is a figure which shows an example of the speech subject data 133 of FIG. It is a figure which shows an example of the value data 134 of FIG. It is a figure which shows an example of the data produced in the sentence structure storage area (FIG. 4A) of a value accompanying operation | movement (FIG. 8) of the importance degree calculation means 1213 of a value. It is a figure which shows an example of the dictionary 135 for sentence structure analysis of FIG.

A. Overview

Hereinafter, specific configuration examples of the opinion collection system and the opinion collection method according to the present embodiment will be described.

Past utterances of a certain utterance subject can be obtained by document retrieval means. Therefore, if there is a means for calculating the degree of importance of a given value from a given text set, the degree of importance of the speaking subject on the value can be calculated. In this embodiment, the means for calculating the importance level for the values of the text set using the text set and the values as arguments is referred to as “importance level calculating means”.
When estimating the importance of the values of the text set, the words given as the values may appear as they are, but it is necessary to consider expressions that are different or highly semantically similar. For example, in the case of medical progress, “innovation of medical technology” and “development of new drugs” are also highly similar expressions. Semantic similarity between values and other expressions is called semantic similarity.
Furthermore, depending on the context in which the same expression appears, the degree to which it supports values also changes. If “medical progress” appears in XX in the context of “XX is important,” support for medical progress is high, but if it is a context such as “XX is not always good.” The support is considered low. The effect of context on support is called “context factor”.
An example of a representative example of this embodiment is as follows. The opinion collection system includes an input unit that inputs a topic and values related to the topic, an opinion collection unit that retrieves an opinion about the topic, a past speech retrieval unit that retrieves a past opinion about the subject of the opinion, It has a calculation means for calculating the degree of importance regarding values and a display unit for displaying retrieved opinions.
In addition, the present embodiment is characterized in that a subject is mapped to a display unit based on a plurality of values, and individual remarks related to a topic are linked to a replay subject and displayed in a form in which the pros and cons regarding the topic are understood. Can do.
The importance calculation means detects expressions having values and semantic similarity from past statements, searches for semantic similarities between the detected expressions and values, and calculates context factors of appearance contexts. The importance level can be calculated by adding a value obtained by merging the two (product or the like).

B. Embodiment

1. System and apparatus

An opinion collection system according to the first embodiment of the present invention will be described. First, the basic configuration of the present embodiment will be described with reference to FIG.

FIG. 1 is a configuration diagram of an opinion collection system 1000 according to an embodiment of the present invention. The opinion collection system 1000 includes an opinion collection device 100 and an opinion browsing analysis support terminal 200, which are connected by a communication network 300. The opinion collection device 100 and the opinion browsing analysis support terminal 200 can be integrated. Printing means 400 such as a printer is connected to the communication network 300 or the opinion browsing analysis support terminal 200. The opinion collection system 1000 is also appropriately connected to other terminals and servers of related departments or terminals and servers of related departments of external organizations via the communication network 300.
The opinion collection device 100 can be configured by a computer including a calculation unit (CPU) 110, a main storage unit 120, an auxiliary storage unit (databases) 130, an input unit 140, a display unit 150, and a communication unit 160. Each unit described below is realized by the unit 110 executing various programs stored in the main storage unit 120. That is, the calculation unit 110 controls the operation of the opinion collection device 100 by executing a program stored in the main storage unit 120.

The main storage unit 120 stores an opinion collection management unit 121 that is a program in which the opinion collection function provided by the opinion collection device 100 is implemented. The opinion collection management unit 121 includes, as constituent elements, an opinion collection unit 1211 relating to a topic, a past statement collection unit 1212 of a utterance subject, a sense of importance calculation unit 1213, and a context factor calculation unit 1214. Further, the main storage unit 120 has an opinion collection work area 1215 for temporarily storing data when executing these means, and various rules to be referred to at the time of execution and opinions collection rules / data collections. A data collection 1216 is also included. Details of the operation of the opinion collection management unit will be described in detail later.
The main storage unit 120 also stores the following means that are called when the opinion collection management unit 121 is executed. That is, the document search means 122, the document entity acquisition means 123, the sentence division means 124, the sentence structure analysis means 125, the word division / part of speech assignment means 126, the name identification means 127, the specific expression extraction means 128, and the like.
Since each of these means 121 to 128 is a known technique, detailed description is omitted. A known method may be used for the document search means 122 and the word segmentation / part of speech assignment means 126. The name identification means 127 is a technique for reducing a plurality of names of the same target (person, place name, book name, etc.) due to subtle differences in notation to a single name. A known technique can also be used for the specific expression extracting unit 128. For example, the sentence dividing means 124 includes a method in which the word division / part-of-speech giving means is applied to a text, and a portion (such as a punctuation mark) recognized as a sentence end is recognized as a break and divided. Non-Patent Document 1 also describes a method. A well-known method can be used for the sentence structure analyzing unit 125.

The auxiliary storage unit (database) 130 is configured by a hard disk or the like, and stores data, a knowledge database such as a dictionary, and the like necessary for executing each means of the opinion collection management unit 121. That is, the document data 131, the search index data 132, the speech subject data 133, the value data 134, the sentence structure analysis dictionary 135, and the like. The document data 131 is data obtained by digitizing a document group describing past opinions together with bibliographic data such as a speaker, date, time, and location. An example of the document data 131 is shown in FIG. 15A. The document search index data 132 is index data used to search the document data 131 at high speed. A part of these data may be stored in a database of an external information processing apparatus connected to the opinion collection apparatus 100 via the communication network 300. An example of the search index data 132 is shown in FIG. 15B. An example of the speech subject data 133 is shown in FIG. 15C, and an example of the value data 134 is shown in 15D. An example of the sentence structure analysis dictionary 135 is shown in FIG. The description of these figures will be described later.
The input unit 140 is a device that receives an operation input from a user, such as a mouse or a keyboard. The display unit 150 displays a screen used when the system administrator or the user operates the opinion collection apparatus 100. The communication unit 160 communicates with the opinion browsing analysis support terminal 200 via the communication network 300, and transmits opinion data, a speech subject data, and a speech subject × value-oriented importance degree data described later to the opinion browsing analysis support terminal 200.

The opinion browsing analysis support terminal 200 includes a calculation unit 210, a main storage unit 220, an auxiliary storage unit 230, an input unit 240, a display unit 250, and a communication unit 260. The calculation unit 210 controls the operation of the opinion browsing analysis support terminal 200 by executing a program stored in the main storage unit 220. The main storage unit 220 stores an opinion browsing analysis support management unit 221 that executes an opinion browsing analysis support function provided by the opinion browsing analysis support terminal 200. The management unit includes an opinion browsing analysis support means 2211 which is a program for implementing an opinion browsing analysis support function and a work area 2212 for storing data generated at the time of execution.
The opinion browsing analysis support management unit 221 displays an opinion browsing analysis support screen 251 on the display unit 250 by using the below-described opinion data, the speech subject data, and the speech subject × value-oriented importance data received from the opinion collection device 100. Let The user uses the opinion browsing analysis support screen 251 to perform operations such as browsing and analyzing opinions.
The input unit 240 is a device that receives an operation input from a user such as a mouse or a keyboard. The communication unit 260 communicates with the opinion collection device 100 via the communication network 300.

FIG. 2 is a diagram illustrating a screen example of the opinion browsing analysis support screen 251 displayed on the display unit 250 of the opinion browsing analysis support terminal 200 of FIG. The computing unit 210 of the terminal provides the opinion browsing analysis support screen 251 on the display unit 250 by executing the opinion browsing analysis support means 2211.
The opinion browsing analysis support screen 251 is displayed when an agenda setting unit 2511, a values setting unit 2512, an option setting unit 2513, an opinion collection instruction unit 2514, an opinion list display unit 2515, and a mouse cursor is placed on each opinion. The individual opinion detail display section 2516 is included. The selection column for affirmation and denial on the right side of the topic setting section is for making a setting to turn on a switch that reverses the approval or disapproval of opinion when negative is selected. For example, if the subject is “prohibit animal experiments”, select “No”.

The opinion browsing analysis support unit 2211 stores the topic input to the topic setting unit 2511 in the topic storage area 221001 in the main storage unit 220, and the values set in the value setting unit 2512 are stored in the value storage area 221002. Stored in
When an opinion collection execution instruction is given from the opinion collection instruction unit 2514, both the topic and values stored in the work area 2212 are transmitted to the opinion collection device 100. From the opinion collection device 100, first, the opinion collection result (detailed later with reference to FIG. 3B) and the data related to the list of comment actors are received, and the data are stored in the opinion data storage area 221003 of the work area 2212 and the comment subject data. Stored in area 221004. Furthermore, the importance data according to the speech subject × value (which will be described in detail later with reference to FIG. 3D) is received from the opinion collection device 100 and stored in the importance data storage area 221005 of the speech subject × value. The opinion browsing analysis support unit 2211 displays the opinion list display unit 2515 based on the contents of the opinion data, the speech subject data, and the speech subject × value-oriented importance data in the work area 2212.

In this figure, “the right or wrong of animal experiments” is selected as the topic, “medical progress” and “life of animals” are selected as related values, and the opinion list display unit 2515 has a horizontal axis indicating the importance of medical progress, and the vertical axis The importance of the life of animals is taken into account, and the speakers (here, A, B, C, D) are arranged with the ordinate and abscissa determined according to each importance, and the pros and cons of each speaker (in this example, the pros and cons) Is displayed with an opinion on the topic.
Although not shown in the figure, when three or more values are set, a setting unit for instructing which and which are set as the vertical axis and the horizontal axis is also displayed. If there is no instruction to select values, it is preferable to select two values that are as independent as possible. Therefore, using an independence index such as a chi-square test between the distributions of the importance levels of individual speakers. Choosing the largest pair is one of the preferred methods.

Since a candidate presentation button is drawn on the right side of the value setting unit 2512, this value recommendation function will be described. In this example, it is assumed that the values are set based on the insight of the opinion collector, but in some cases it may be desirable to recommend from the system side. It is a function. An example of an implementation method is shown below.
Documents relating to the topic are searched, and all the values registered in the value data 134 are searched for all words (see the explanation part of FIG. 9A described later) that are judged to have semantic similarities thereto. Count and sum the frequency of occurrences in the document. Values are sorted in descending order of the frequency total count, and a preset number is selected from the top (for example, 5). Assuming that the selected values are set, the data of importance of the utterance subject × value is created. A pair of values is listed in descending order of independence in accordance with the method for automatically selecting values versus when three or more values are set. A method of recommending this higher rank as a candidate can be considered.

FIG. 3A is a diagram showing a configuration of value data stored in the value storage area 221002 in FIG. The value data includes a local number for distinguishing different values (only 1 and 2 in the example of FIG. 2), a value identifier, and its contents. The identifier of values is a value in which identifiers are entered for values already registered in the values data 134.
FIG. 15D is a diagram illustrating a data configuration of the value data 134. The value data includes the value identifier corresponding to the value identifier. In this example, “medical progress” is already registered as the 086 value.

FIG. 3B is a diagram showing a configuration of opinion data stored in the opinion data storage area 221003 in FIG. In addition to the content of the opinion, the data indicates whether the content is in favor / support (+1) or opposite / negative (-1) for the topic, and the strength of the relevance to the topic The relevance score, the reason evidence score indicating whether or not the reason for the opinion and the evidence are shown, the content of the opinion, the ID of the document in which the opinion is expressed, the title, and the speaking subject identifier are included.

FIG. 3C is a diagram illustrating a configuration of data relating to a speech subject stored in the speech subject data storage area of FIG. 2. The data includes a local number for numbering a speech subject, a speech subject identifier registered in the speech subject data 133, a name, a belonging organization (or higher organization) identifier, and the like.
FIG. 15C is a diagram showing a data structure of the speech subject data 133. The speech subject data is composed of a speech subject name corresponding to the speech subject identifier, an alias, a belonging organization (or higher organization) identifier, and the like. The name is required but the others are optional.

FIG. 3D is a diagram illustrating a configuration of the speech subject × value-oriented importance data stored in the comment subject × value-oriented importance data storage area 221005 in FIG. 2. In addition, "x"
Represents a matrix. The data is obtained with respect to a pair of each local number (described in FIG. 3C) of the collected voice of the opinion and each local number (described in FIG. 3A) stored in the values storage area. This is tabular data in which the degree of importance of the speaking subject with respect to the values is described.

FIG. 4A is a configuration diagram of the opinion collection work area 1215 of the opinion collection device 100. This work area includes a topic storage area 121501, a values storage area 1215002, an opinion data storage area 1215003, a speech subject data storage area 1215004, a speech subject × value-oriented importance storage area 1215005 (the above is a data storage area of the terminal side work area) 2212001 to 221005), a search condition storage area 1210501, a search result storage area 1215011, a document entity storage area 1215012, a bibliographic information storage area 1215013, a sentence storage area 12105020, a sentence structure storage area 1215021 of values, The area includes a reverse flag storage area 1215022, a sentence structure storage area 1215023, a context factor calculation work area 12105024, and the like.

FIG. 4B is a configuration diagram of the opinion collection rule / data collection 1216 of the opinion collection apparatus 100. The rule / data collection includes approval / disagreement expression data 1216001, reason / evidence expression data 1216002, negative expression data 1216003, statement expression data 1216004, semantic similarity word pair data 1216011, promotion / suppression phrase data 12116012, main factor data 1216021 , Auxiliary part factor data 1216022, modification part factor data 1216023, and the like.

2. processing

FIG. 5 is a diagram illustrating a sequence in which the opinion browsing analysis support management unit 221 and the opinion collection management unit 121 in FIG. 1 operate via the communication network 300. Hereafter, each step (step F1-F10B) of FIG. 5 is demonstrated.

(Steps F1 to F3):
The opinion browsing analysis support terminal 200 activates the opinion browsing analysis support screen 251, and the topic and values (plurality) are set by the user's input operation, and options are specified. An opinion collection execution request is sent to the opinion collection management unit 121 of the opinion collection apparatus 100 with an option added to the topic and values.

(Steps F4 to F6):
The opinion collection means 1211 of the opinion collection device 100 collects opinions on the topic set and transmitted by the opinion browsing analysis support terminal 200 from the document data 131, and the opinion data (D1) whose configuration example is shown in FIG. 3B (Details will be described later). The opinion collection device 100 stores the opinion data in the opinion data storage area 1215003 on the opinion collection device 100 side in step F4. Further, the opinion collection device 100 transmits the opinion data to the terminal side in step F5, and the opinion browsing analysis support terminal 200 stores the opinion data in the opinion data storage area 221003 on the terminal side.

(Steps F7 to F10):
The opinion collection device 100 further collects each of the comment subjects included in the opinion data (D1), creates the comment subject data (D2), the comment subjects that constitute the opinion data (D1), and the set values. On the other hand, the past degree collecting means 1212 and the importance level calculation means 1213 of the speech subject are used to calculate the importance level regarding the values of the speech subject, and the obtained importance levels are summarized in a table as the talk subject × Value-oriented importance data (D3) is created. The opinion collection device 100 stores the obtained data D2 and D3 in the speech subject data storage area 1215004 and the speech subject × value-oriented importance data storage area 1215005 on the opinion collection device 100 side, respectively. Further, the opinion collection device 100 sends the obtained data D2 and D3 to the terminal side in step F9, and the opinion browsing analysis support terminal 200 sends D2 to the terminal-side speech subject data storage area 221004, D3 on the terminal side. The data is stored in the importance data storage area 221005 of the utterance subject × value.

(Step F11A / B):
When the result display option is the normal display, the opinion browsing analysis support terminal 200 displays the result based on the opinion data (D1). In the case where the result display option is a display reflecting the importance level of the value of the speech subject, the opinion browsing analysis support terminal 200 assigns each of the comment subjects to the value based on the importance data of the speech subject × value (D2). It is displayed at the coordinate position corresponding to the degree, and the opinion of each utterance subject is displayed at that position.
When the values are “medical progress” and “animal life”, the opinion browsing analysis support terminal 200 takes the importance of medical progress on the horizontal axis and the importance of animal life on the vertical axis. The ordinate and abscissa are determined and arranged according to the importance of each person's sense of values, and each speaker's approval / disapproval opinion is displayed in that position, and the approval opinion is indicated by ○ and the disagreement is indicated by ×.

FIG. 6 is a diagram illustrating a procedure in which the opinion collection unit 1211 collects opinions related to a topic from a given topic in step F4 of FIG.
In process 1211001, the opinion collection unit 1211 creates a search condition Q for a given topic P. For example, the search condition Q is obtained by applying the word division / part-of-speech giving means 126 to the topic P and ORing the list of content words obtained by removing function words such as particles and auxiliary verbs. It should be noted that a simple OR combination includes a case where a plurality of content words appear at positions far away from each other. Therefore, in order to prevent this, it is a good method to add a condition for limiting the distance between the appearance positions.
In process 1211002, the opinion collection unit 1211 gives the search condition Q to the document search unit 122. The search means executes a search using the search index data 132 to obtain a list of document IDs satisfying the search condition Q, and stores it in the search result storage area 1215011.

FIG. 15B is a diagram showing a data structure of the search index data 132. The search index data 132 includes a document ID and a list of index words (index data). The index word describes the number of appearances in the document, a list of appearance positions, and the like. The search index data 132 is created in advance by extracting index words based on the document data 131. Further, in addition to the index word appearing in the text, an index related to the speaking subject is included in the data in order to perform a search specifying the speaking subject later in the processing 1213002 in FIG. 8 (“speaking subject = W024 ").
As for index terms appearing in the text, only the index terms may be used if the search formula is only a logical combination of search terms (a combination of AND, OR, NOT, etc.). If the priority is raised or lowered depending on the number of appearances, enter the number of times. In addition, in order to restrict how many search terms appear, it is necessary to describe the appearance position. Although the present embodiment can be implemented using only index words, in order to search document data having a higher relevance to the topic in a higher order, it is better to have information on the number of appearances and the appearance position. In response to an OR join search request, the document search means 122 accesses this data and outputs a list of document IDs of documents including any word in the conditional expression as an index word.
In a loop 1211003, the opinion collection unit 1211 performs the following processing 121110 to 1211013 on each retrieved document ID (I).
In processing 12111010, the opinion collection unit 1211 refers to the document data 131 for the given document ID (I) by the document entity acquisition unit 123, and the document content text corresponding to the document ID (I) ( Body) and title. The opinion collection unit 1211 stores the document ID, the text (text) of the document content, the title, and the like in the opinion data storage area 1215003.

FIG. 15A shows the data structure of the document data 131. The document data 131 may include information about a date subject identifier and date / time for correspondence with later-described message subject data 133 in addition to the document title and body in a form corresponding to the document ID. The essential component of these is the body. If there is no title, the head part of the text can be substituted.
In processing 1211011, the opinion collecting unit 1211 divides the text into sentence units by the sentence dividing unit 124.
As the sentence dividing means, there is a method in which after applying the word dividing / part-of-speech giving means 126 to text, a part (such as a punctuation mark) recognized as a sentence end is recognized as a break and divided.
In a loop 12111012, the opinion collection unit 1211 performs the following processing 1211020 to 1211023 for each of the divided sentences (S).
In processing 12111020, the opinion collection unit 1211 calculates the relevance R between the sentence S and the topic P. The opinion collection unit 1211 stores the relevance R in the opinion data storage area 1215003. The relevance R is a value indicating a higher value as more types of content words constituting the topic appear in a narrower range.

An example of the calculation method is shown below.
When N types of content words constituting the topic are included in the sentence S, the number of words in the minimum interval in which the J types of content words are included in the sentence S for each J, where J is an integer from 1 to N ( R (J) = J × J ÷ D (J) where D (J) is the number of words from the first appearance position to the last appearance position). One of the good methods is to set the relevance R to the maximum value of r (J) when J is moved from 1 to N. If N types of content words appear continuously, D (N) = N, and r (N) = N × N ÷ N = N, that is, the number of types of content words. The farther away, the lower the value.
Note that when N (K> N / 2) appear continuously and the rest appear at very distant positions, r (K) = K, but if J> K Since D (J) increases, it is considered that r (J) <K. In this case, the relevance R is considered to be K.

The relevance calculation method will be described with a specific example. Here, it is assumed that A to Z represent some word, and the sentence S is “ABCDE FGAGBAC”. The content words of the topic P are A, B, F, and K. In this case, since sentences S include A, B, and F, N = 3 types of content words are included. Therefore, in order to calculate the relevance R of the sentence S to the topic P, r (J) is calculated for J = 1, 2, 3, and the maximum value is obtained. When J = 1, it is the shortest interval in which one type of word appears, and since it is always its own 1, D (1) = 1 and r (1) = 1 × 1 ÷ 1 = 1 are calculated. . In the case of J = 2, D (2) is 2 for the shortest interval between A and B, 3 for the FGA for the shortest interval for A and F (no matter the order), and 4 for the shortest interval between B and F is FGAB. is there. Accordingly, the shortest interval between the two types of content words is 2 in the case of A and B, so D (2) = 2, and r (2) is 2 × 2 ÷ D (2) = 2. When J = 3, the three types of words are only combinations of A, B, and F. Since the shortest interval in which A, B, and F all appear is FGAB of 4, D (3) = 4 and r (3) = 3 × 3 ÷ D (3) = 2.25. Eventually, the maximum value of r (J) is 2.25 when J = 3. Assuming that XYZ is inserted between G of the sentence S and the next A to be “A BCD E F G X Y Z A B C D”, r (1) r (2) is the same. In D (3), the shortest interval in which all of A, B, and F appear is 6 of the first ABCDEF, so r (3) = 3 × 3 ÷ 6 = 1.5. In this case, r (2) is calculated as relevance R = 2 at maximum. That is, even if the number of types is large, if they appear apart, a portion where a smaller number of types of words appear compactly may win.
Here, for simplicity, r (J) = J × (J ÷ D (J)). However, when (J ÷ D (J)) is directly multiplied by J, J types of the shortest appearance section length (D There is a concern that the effect of (J)) is too strong, and in order to prevent this, it is a good method to take the route (0.5th power) of (J ÷ D (J)) and then multiply by J. In addition, when a threshold value D0 relating to the section length is provided and the difference due to the section length is not taken into account when it appears within the section length, J1 = MAX (D0, J), D1 = MAX (D0 , D (J)) and r (J) = J × (J1 ÷ D1) is a good method.

In the process 12111021, the opinion collecting unit 1211 refers to the approval / disapproval expression data 121601 (FIG. 7A) and the negative expression data 1216003 (FIG. 7C) for the sentence S related to the topic P, and supports / disagrees with the topic. Determine. Whether or not there is a relationship can be determined, for example, by comparing the relationship R with a predetermined threshold value. The opinion collection unit 1211 stores items (+1, −1) for approval and disagreement in the opinion data storage area 1215003. If there is an expression of approval and there is no expression of approval, the expression is approval, and conversely, if there is an expression of expression, it is determined to be objection, and if there are multiple negative expressions, it is reversed if the number is odd.
In the processing 1211022, the opinion collecting unit 1211 refers to the reason / evidence expression data 1216002 (FIG. 7B) for the sentence S that is related to the topic P and is determined to be in favor or not, and determines whether the reason / evidence exists. judge. The opinion collection unit 1211 stores the reason evidence score in the opinion data storage area 1215003 depending on the presence / absence of evidence.
In process 1211023, for these sentences S, the opinion collection unit 1211 identifies the speaking subject by collating with the speech expression data 1216004 (FIG. 7D). The opinion collection unit 1211 stores the comment subject identifier in the opinion data storage area 1215003. For example, it is possible to specify a speaking subject assuming that a syntax element corresponding to the subject of the speech expression corresponds to the speaking subject. If the name collating unit 127 can return to the entry of the speech subject data 133, the entry is taken as the speech subject. When the date and time of the utterance can be specified by the specific expression extracting means 128, the date and time is taken as the date and time of the utterance.

FIG. 15C shows an example of the speech subject data 133. As described above, the speech subject data is composed of a name, an alternative name, a belonging organization (or higher organization) identifier, and the like. When the predicate of the speech expression shown in FIG. 7D appears before or after the sentence S, the opinion collection unit 1211 takes the subject and matches the name or the alias of the speech subject data. Identify the subject. If the document cannot be identified, the opinion collection unit 1211 estimates that the speaking subject of the document is registered in the document data (FIG. 15A), as will be described later.
Regarding the identification of the date and time, the opinion collection means 1211 matches the pattern such as “<number> year <number> month <number> day” with the context before and after, and if it succeeds, Take as date and time. When the matching cannot be obtained, the opinion collecting unit 1211 estimates the date / time of the document if the date / time of the document is registered in the document data (FIG. 15A).

The opinion collection means 1211 uses the default value for those in which the statement subject and the statement date / time cannot be specified after exiting the loop 12111012 by repeating the processing from 12121020 to 12111023. That is, in the process 12111013, if there is at least one sentence that is relevant to the topic P and can be approved / disagreeed, the default speaking subject, the default speaking date and time, and the title for this document (I) Obtained from data 131 (FIG. 15A).
The opinion collection means 1211 exits the loop 1211003 by repeating the processes from 12111010 to 12111013, and then in the process 1211004, the obtained opinion data (FIG. 3B) is used as the relevance R (relevance score) with the topic P. Are sorted in descending order as the first sort key, and then the reason / evidence presence / absence (reason / evidence score) is sorted in descending order as the second key. If the number of opinions collected exceeds the upper limit specified by the option, the part whose rank after sorting exceeds that number is discarded.

FIG. 7A is a diagram showing an example of approval / disagreement expression data 121001 used for collecting opinions. Words and pros and cons information are written. As for approval or disapproval, here, the approval is 1 and the opposite is -1.
FIG. 7B is a diagram showing an example of reason / evidence expression data 1216002 used when collecting opinions. Includes words and grammatical information used to show reasons and evidence.
FIG. 7C is a diagram showing an example of negative expression data 1216003 used for collecting opinions. Contains words and grammar information used to express negation.
FIG. 7D is a diagram illustrating an example of the utterance expression data 1216004 used when collecting opinions. Contains words and grammar information used to express a statement.
7A to 7D, the left figure shows the Japanese version, and the right figure shows the English version.

In FIG. 8, the importance level calculation means 1213 collects the past statements of the speaking entity for the given speaking entity and the given values, and calculates the importance regarding the values from the collected statements. It is a figure explaining a procedure.
First, in processing 1213000, the importance level calculation means 1213 clears the importance level value (V) to be obtained to zero.
In process 1213001, the importance level calculation means 1213 applies the sentence structure analysis means 125 to a given value and stores the obtained sentence structure (VS) in the value sentence structure storage area 1215021. When the main part of the topmost syntax element of the syntax structure corresponds to the promotion / suppression word by collation with the promotion / suppression word data 12116012 (FIG. 9B), the syntax element to be promoted / suppressed is the highest order. In the case of suppression, the value inversion flag (Rev) is turned on. The initial value of Rev is off.

FIG. 16 shows an example of data stored in the sentence structure storage area 1215021 of values when the values are “medical progress”. In the initial structure, which is the output result of the sentence structure analysis means, the top (first syntax element) is a simple sentence whose main part is the predicate “advance”, and the second syntax element that is the main subject of the operation is the main part. It is a noun "medicine". Referring to the promotion / suppression word / phrase data 12116012 (FIG. 9B), it can be seen that “advancement” has a facilitating property with respect to the action subject. Therefore, the second syntax element (main part is “medicine” ") Is taken as the top-level syntax element. Since “advance” is a promotion direction, the inversion flag is turned off.
The sentence structure analyzing unit 125 constructs a sentence structure while referring to the sentence structure analyzing dictionary 135.

FIG. 17 shows an example of the dictionary. In addition to the word name and its part-of-speech information, the sentence structure analysis dictionary 135 includes, in the case of a word representing an action, a semantic role such as an action main and an action target, and a search rule for a term that plays the meaning role. For example, in the case of progress, there is a subject as a semantic role, but the search rule is the main actor. In the search rule list attached to the dictionary, the priority order of the particles when searching for the actor is shown, and the corresponding terms are searched in that order. In the case of “medical progress”, since the fourth “no” corresponds, medicine is taken as an operation. However, when an auxiliary verb such as “advance” is attached, a term is searched according to the rule of action.
In processing 1213002, the importance level calculation unit 1213 gives the search condition to the document search unit 122 that the message is from the given message subject, and executes a search based on the search index data 132, thereby executing A list of document IDs that are statements is obtained.
In processing 1213003, the importance calculation means 1213 refers to the affiliation (higher order) organization identifier of the speech subject data 133 (FIG. 15C) as a measure when the list of document IDs does not reach a predetermined number. If it is registered, a search is executed on the condition that the speaking subject is the higher organization, and a list of document IDs is obtained to compensate for the shortage. This processing is not necessarily essential, but is effective as a remedy when the number of documents to be searched is small. In addition, since this process may have a side effect, it is desirable to be able to set as an option whether or not to do so.
Subsequently, in a loop 1213004, the importance level calculation unit 1213 executes processing from processing 1213011 to processing 1213013 for each searched document ID (I).
In processing 1213011, the importance level calculation unit 1213 gives the document ID (I) to the document entity acquisition unit 123, and acquires the text (text) of the document content from the document data 131 (FIG. 15A).
In process 1213012, the importance calculation means 1213 divides the text into sentence units by the sentence dividing means 124.

Subsequently, the processing enters a loop 1213013, and the importance calculation means 1213 performs processing 1213021 and loop 1213022 on each of the divided sentences S.
In process 1213021, the importance level calculation means 1213 applies the sentence S to the sentence structure analysis means 125, obtains syntax structure data, and stores it in the sentence structure storage area 1215023. Later, in the explanation part of FIG. 10, the syntax structure creation method will be described in detail using example sentences. The sentence structure analysis is a process based on the grammar and syntax analysis dictionary 135 (FIG. 17). Next, entering the loop 1213022, the importance calculation means 1213 performs the processing 1213031-1213033 for the syntax elements P constituting the syntax structure having values and semantic similarity.
In process 1213031, the importance calculation means 1213 calculates the sentence structure (VS) of values and the semantic similarity Sim of the syntax element P. Here, the VS is obtained by applying the sentence structure analyzing means 125 to a given value in the processing 1213001, and is stored in the sentence structure storage area 1215021 of the value.

Details of the method of calculating the semantic similarity Sim will be described below.
When the main part of the top syntax element of the sentence structure VS matches the main part of the syntax element P, or whether it corresponds to the similar phrase pair registered in the semantic similarity phrase pair data 12116011 and matches Is Sim = 1, and in the case of a similarity pair, the similarity value written in the data is Sim. If the values are only the top syntax, this is the end. When the value is accompanied by a term, if there is no term of the same role child on the side of the syntax element P, Sim = 0 is obtained. If there is a term, the similarity between the terms is calculated and multiplied by the Sim. If there are multiple terms on the values side, it is a condition that there is a term of the same role child on the syntax element P side for all, and all the similarities between the corresponding terms of the sentence structure VS and the syntax element P are all Calculate and multiply. In the case where the value modification side has a linkage modification, it is a condition that the syntax element P side also has a corresponding linkage modification, and the similarity between them is calculated and multiplied.

In the case of the syntactic structure of the example sentence shown in FIG. 10 below, if the value is “medical progress”, the syntactic structure VS of values is mainly a medical term, so the value of Sim is positive. This means that the syntax element is the 10th term. In this case, since the main word matches, Sim = 1.0. If the main part of the syntax element 10 is medical, the similarity coefficient with “medicine” in the semantic similarity phrase pair data 1216011 is 1.0, so that Sim = 1.0, and the main part is “drug”. For example, since the similarity coefficient is 0.2, Sim = 0.2.
In process 1213032, the importance calculation means 1213 calculates the context factor CtxFactor in the sentence S of the syntax element P by the context factor calculation means 1214. Details of the calculation method will be described later with reference to FIGS.
In process 1213033, the importance level calculating means 1213 calculates the value support (s) of the syntax element P from the semantic similarity Sim and the context factor CtxFactor, and adds it to the value (V) of the importance to be obtained. . If the value inversion flag (Rev) is on, the value is subtracted. As a method for calculating the degree of support for values (s), it is one preferable method to obtain the value by the product of the similarity level Sim and the context factor CtxFactor.
As described above, the triple loops 1213022, 1213013, and 1213004 are exited, and the importance level calculation unit 1213 outputs the value of the importance level V obtained in the processing 1213005.

FIG. 9A is a diagram illustrating an example of semantic similarity phrase pair data 1216011 used for importance calculation. The data includes word / phrase pairs having semantic similarity and their similarity coefficients. The similarity coefficient is a real number greater than 0 and less than or equal to 1.0. The larger the value, the more similar.
FIG. 9B is a diagram illustrating an example of the promotion / suppression word / phrase data 12116012 used when calculating the importance level. It includes a phrase having an accelerating property or an inhibiting property, a target role child that indicates what it promotes or inhibits, and a coefficient that indicates the degree of promotion or inhibition. There are generally a plurality of target role children, which are listed in order of priority. For example, in the case of “promotion”, if there is a syntax element “XX” corresponding to the target, it is taken as the target of promotion, and there is no syntax element that indicates the subject of action “XX” If an element is present, it indicates that it is a promotion target. The promotion / suppression coefficient is positive when it is positive and negative when it is negative.

FIG. 10 shows an example of the sentence stored in the sentence storage area 12105020 and the result of analyzing the syntax structure when the importance is calculated using the method shown in FIG. It is the figure which showed the content. The example sentence is "I wonder if we need to think carefully what we should do in order to remove the factors that impede the development of medicine one by one." For reference, English example sentences are also shown. In the structure of the example sentence, the highest-level syntax element (element number 1) is a single sentence in which the predicate as the main part is “think”, and the action target is the number 2 syntax element. . In addition, it is also described that the auxiliary part at the end of the sentence is “... I wonder if it is necessary to try”, and the modifier (continuous modification) is “slowly”.
The sentence ending auxiliary part is obtained by taking only the expressions registered in the auxiliary part factor data 1216022 as shown in FIG. 12B and the concatenation of function words such as auxiliary verbs, particles, and conjunctions from the sentence ending. In the case of this example, the three parts of “Try”, “Need”, “Nease”, “Neka” are registered in the auxiliary part factor data. The remaining “no”, “de”, and “ha” are formal nouns, auxiliary verbs, and particles, respectively, which are taken as function words. Next, “think” is taken from the syntax analysis dictionary (FIG. 17) as a main part that takes the object (thinking contents) and the subject (who thought) as semantic roles. As a target, the section from the top to “what to do” is taken according to the section search rule, and the subject is not applicable. Also, adverbs that modify the main part, such as “carefully”, are added to the modifier.

Next, the analysis proceeds to the part corresponding to the content to be considered, that is, the part from the beginning of the sentence to “what to do”. Since this part matches the pattern of purpose (A) -means (B) of “do B for A”, the second syntax element is a compound sentence whose main part is “purpose-means” and is in role 1 Means and role 2 are registered with means. In this example, the syntax elements are No. 6 and No. 3, respectively.
The part corresponding to the means is “what we should do”, and the function words “should” and “ka” are taken from the end of the sentence to become the end of sentence auxiliary part. Next, it can be seen from the lexicon for parsing that the verb “do” takes the subject and the object as semantic roles, and “we” and “what”, respectively, according to the search rules of the corresponding terms. From the above, the third syntax element is a simple statement whose main part predicate is “Yes”, “We” corresponding to the subject is the fourth syntax element, and “What” corresponding to the target is the fifth syntax element. It becomes.
In the syntax element No. 6 corresponding to the purpose, the predicate “remove” is a simple sentence of the main part, the object to be removed is the syntax element of No. 7, and the noun “factor” is the term of the main part. This 7th term has a linkage modification, which is the 8th syntax element whose main part is the predicate “inhibit”. The No. 8 operator is the No. 7 factor, and the obstructed target is the No. 9 syntax element. The ninth syntax element is a simple sentence having the predicate “development” in its main part, and the developing subject is a term having the tenth medicine in its main part. Since the English sentence analysis method is performed in the same manner, the description is omitted.

FIG. 11 is a diagram illustrating the contents of the context factor calculation work area 12105024 when the importance level is calculated for the same example sentence as FIG. The syntax element number corresponds to the syntax element number in FIG. For each syntax element, a main factor column, an auxiliary factor factor, a modifier factor column, and a context factor calculation column are facilitated.
In the main factor column, a calculation formula corresponding to the main part of the syntax element is stored with reference to the main factor data 1216021. The value calculated by referring to the auxiliary part factor data 1216022 is stored in the auxiliary part factor column. In the calculation of the auxiliary part factor, it is a simple and preferable method that the auxiliary part is matched with the expression registered in the data with the longest match from the beginning, and a value is obtained by multiplying the factors of the matching. When the matching cannot be obtained, the default value is set to 1.0. In the figure, the default value is displayed in parentheses.
In the modifier part factor column, a corresponding value is stored if there is a corresponding part with reference to the modifier part data 1216023. If not, the default value is 1.0.
The calculation formula described in the main factor is a calculation formula for calculating the context factor of the syntax element from the context factor of the term (subordinate syntax element). The auxiliary part factor is a factor relating to the auxiliary part of the syntax structure, and the modifier part factor is a factor relating to the modifier part of the syntax structure. The context factor specifies a syntax element and calculates from there up to the top of the syntax structure.

In the case of the example sentence, if the value is “advance in medicine”, the value and the semantic similarity are positive, as shown in the explanation of the process 1213031 in FIG. (The term corresponding to “medicine”). Therefore, here, a method for calculating the context factor of the tenth syntax element will be described.
First, a context factor of 1.0 is given to the 10th syntax element. Next, a syntax element determined only from the 10th syntax element is searched, and it is understood that the 9th syntax element is it. The main factor is calculated as 1.0 with 10th context factor × 1.0. Repeating the same work, (1) No. 8 context factor is -1.0, (2) No. 7 is -1.0, (3) No. 6 is 1.0, (4) No. 2 Is calculated in order such that 1.0 is 1.0 and (5) No. 1 is 0.6. (1) to (5) will be described in detail below.
(1) No. 8 is a syntax element “inhibit OO”, and the main factor is (−1) × (the ninth factor to be inhibited = 1.0). From this, -1.0 is calculated. (2) Since No. 7 has been continuously modified by No. 8, the factor of No. 8 is taken over and becomes -1. (3) No. 6 is “Remove No. 7”, and the main factor is calculated as −1.0 × (No. 7 factor to be removed = −1.0) = 1.0.
(4) No. 2 is (for No. 6 purpose) (No. 3 is performed), and the main factor is calculated by No. 6 and No. 3 Max. In this case, since No. 6 is 1.0 and No. 3 is 0.0, the main factor is calculated as 1.0. (5) For No. 1, the main part is “considering No. 2” and the factor of No. 2 × 0.8 is calculated as 0.8. Further, the auxiliary part factor is calculated to be 0.7 because it includes “Naka Kana” with reference to the auxiliary part factor data 1216022 in FIG. 12A, and the factor of “slowly” is 1 with reference to the modifier part data 1216023. .2 is calculated. By multiplying these, the context factor is calculated as 0.8 × 0.7 × 1.2 = 0.67.

FIG. 12A is a diagram illustrating an example of main factor data used when calculating a context factor. The main factor data includes a type of syntax element (compound sentence / single sentence), a main part, a list of term roles, and a main factor. When the syntactic element type is complex, the number of causes-results and purposes-means is limited. When the syntax element is a simple sentence, the verb is the main part, and the action main and the action target are the role of the term. The main factor is given as a calculation formula that calculates from the value corresponding to the term.
FIG. 12B is a diagram illustrating an example of auxiliary unit factor data used when calculating the context factor. The auxiliary part factor data includes the value of the factor corresponding to the expression of the auxiliary part. The absolute value is larger the more confidently speaking, the smaller the absolute value in the case of a blurred expression. In the case of negative, it becomes a negative value. “No” of No. 3 is included in No. 1 and No. 2, but the longer one has priority, and No. 3 is not applied to the portion where No. 1 or No. 2 matches.
FIG. 12C is a diagram illustrating an example of modifier part factor data used in context factor calculation. Includes modifiers such as adverbs and their factor values. Large values are given to words of work that strengthen, and small values are given to words of work that blur.

As described above, the opinion collection system according to the present embodiment sets a plurality of values related to a topic together with the topic of interest, and how much importance is given to the above-mentioned values by the subject of the opinion on the topic. The display position of the speaking subject is determined based on the value, and the opinion of the speaking subject is presented based on the position. This makes it possible to read and analyze the collected opinions while grasping the background of the idea of how much importance the speaker has regarding the values.

In addition, as shown in FIG. 13, it is considered that the importance of values changes with the age even if the subject is the same, but by providing a setting means unit 2517 for age classification, It is possible to capture how the background related to the values of the speaking subject changes.

As a specific application example, Example 1 is intended for collecting opinions on social issues that are divided into pros and cons, but may also be opinions on products and services, and political, foreign, and security topics. .

FIG. 14A is a diagram when applied to collecting opinions on products and services.
In the case of products and services, values and performance are typical values. It is assumed that there are price-oriented, performance-oriented, and balanced groups depending on the person who expresses the opinion. Opinion list display unit 2515 is a diagram that displays the results with price on the horizontal axis and performance on the vertical axis. Products with good performance but high price tend to be well-received by performance-oriented people and unpopular by price-oriented people. Under such circumstances, if there is a person who evaluates the term with a price-oriented person, he would like to refer to that opinion.
Since the viewpoint of performance varies depending on the product, it is necessary to register words that are considered to have semantic similarity with the performance in the semantic similarity phrase pair data 1216011 (for example, speed, strength, capacity, etc.). , Safety etc.).

On the other hand, FIG. 14B shows a case where a topic in which there is a conflict of opinions among nations is taken up in the case of opinion analysis on a topic on politics, diplomacy, and security.
When Country B is opposed to Country P's assertion P when Country A and Country B are in conflict, it is a scene that analyzes what opinions are coming from other countries. is there. In this case, as values, it can be considered that the horizontal axis indicates the degree of emphasis on the relationship with Country A and the vertical axis indicates the degree of importance on the relationship with Country B. This balance will vary from country to country. Normally, if a person from country B is in favor of opinion P, or if an opinion is against the opinion P from a country in which country A is important, it would be of interest to be surprised.

C. Effects of the embodiment

According to the present embodiment, for a group of documents in which the content of a statement is described, a plurality of opinions regarding a given topic, and a degree of importance of each speaker regarding a plurality of values related to the topic (background, etc.) You will be able to read while grasping. Further, according to the present embodiment, each opinion can be read while keeping in mind unexpectedness and reliability, and the quality and efficiency of opinion analysis can be improved.

D. Appendix

In addition, this invention is not limited to an above-described Example, Various modifications are included. For example, the above-described embodiments have been described in detail for easy understanding of the present invention, and are not necessarily limited to those having all the configurations described. Further, a part of the configuration of one embodiment can be replaced with the configuration of another embodiment, and the configuration of another embodiment can be added to the configuration of one embodiment. Further, it is possible to add, delete, and replace other configurations for a part of the configuration of each embodiment.
Each of the above-described configurations, functions, processing units, processing means, and the like may be realized by hardware by designing a part or all of them with, for example, an integrated circuit. Each of the above-described configurations, functions, and the like may be realized by software by interpreting and executing a program that realizes each function by the processor. Information such as programs, tables, and files for realizing each function can be stored in a memory, a hard disk, a recording device such as an SSD (Solid State Drive), or a recording medium such as an IC card, an SD card, or a DVD.
Further, the control lines and information lines indicate what is considered necessary for the explanation, and not all the control lines and information lines on the product are necessarily shown. Actually, it may be considered that almost all the components are connected to each other.

  An opinion collection apparatus and system and an opinion collection method according to the present invention include an opinion collection program for causing a computer to execute each procedure, a computer-readable recording medium storing the opinion collection program, and an internal memory of the computer including the opinion collection program Can be provided by a program product that can be loaded on the computer, a computer such as a server including the program, and the like.

DESCRIPTION OF SYMBOLS 100: Opinion collection apparatus, 110: Operation part, 120: Main memory part, 121: Opinion collection management part, 1211: Opinion collection means, 1212: Past comment collection means of a speech subject, 1213: Importance calculation means of values 1214: Context factor calculation means, 1215: Opinion collection work area, 1216: Opinion collection rules and data collection, 122: Document search means, 123: Document entity acquisition means, 124: Sentence division means, 125: Sentence structure analysis means, 126: Word division / part of speech adding means, 127: name identification means, 128: specific expression extracting means, 130: auxiliary storage unit, 131: document data, 132: index data for search, 133: speech subject data, 134: value data 135: sentence structure analysis dictionary, 140: input unit, 150: display unit, 160: communication unit.
1215001: Topic storage area, 1215002: Values storage area, 1215003: Opinion data storage area, 1215004: Speaking subject data storage area, 1215005: Speaking subject × value-oriented importance storage area, 1210501: Search condition storage area, 1215011: Search result storage area, 1215012: Document entity storage area, 1215013: Bibliographic information storage area, 12105020: Sentence storage area, 1215021: Sentence structure storage area of values, 1215022: Reverse flag storage area of values, 12125023: Sentence structure storage Area, 121024: Context factor calculation work area.
1216001: Approval / opposite expression data, 1216002: Reason / evidence expression data, 1216003: Negative expression data, 1216004: Speech expression data, 1216011: Semantic similarity phrase data, 12116012: Promotion / suppression phrase data, 1216021: Main factor data 1216022: Auxiliary part factor data, 1216023: Modifier part factor data.
200: Opinion browsing analysis support terminal, 210: Calculation unit, 220: Main storage unit, 221: Opinion browsing analysis support management unit, 221: Opinion browsing analysis support management unit, 2211: Opinion browsing analysis support means, 2211: Opinion browsing analysis Means 2212: work area 230: auxiliary storage unit 240: input unit 250: display unit 251: opinion browsing analysis support screen 2511: topic setting unit 2512: value setting unit 2513: option setting unit 2514: Opinion list display unit, 2515: Opinion list display unit, 2516: Detailed display unit of individual opinions, 2517: Age division setting unit, 260: Communication unit.
221001: Topic storage area, 221002: Values storage area, 221003: Opinion data storage area, 221004: Talking subject data storage area, 221005: Talking subject × value-oriented importance data storage area.
300: communication network, 400: printing means, 1000: opinion collection system,

Claims (15)

An opinion collecting device,
A plurality of document data including the document content in advance, a plurality of opinion data including the document content and the speaking subject, a storage unit holding importance data for each speaking subject and each value,
An arithmetic unit;
With

The computing unit is
Receiving a topic entered by the terminal that defines what documents to collect and one or more values that are considered to have an impact when determining the pros and cons of the topic;
Search the document data for the document content related to the received topic,
Finding a set of speaking subjects of the retrieved document content, storing a plurality of opinion data including the document content and the speaking subject in the storage unit,
For each speaking subject included in the opinion data, calculate the importance for each of the values,
Create importance level data for each subject and value from the calculated importance level, store it in the storage unit,
An opinion collecting apparatus, wherein the importance data classified by each speaking subject and each value is displayed on a display unit or output to an output unit.
In the opinion collecting device according to claim 1,
The opinion data further includes a relevance to the topic,
The computing unit is
Seeking a set of speaking subjects of the retrieved document content and a relationship indicating the strength of the relationship with the topic;
An opinion collecting apparatus that creates a predetermined number of the plurality of opinion data selected in descending order of relevance.
In the opinion collecting device according to claim 1,
The document data further includes a document content corresponding to the document ID. The opinion data further includes a document ID and a relevance to the topic.

The computing unit is
The topic is divided into words, and a search condition related to the topic is created from a list of content words obtained by removing particles, auxiliary verbs, and other function words, searching the document data, and satisfying the search condition Get a list of document IDs,
For each retrieved document ID, refer to the document data to obtain the document content corresponding to the document ID;
For each sentence in which the document content is divided, the relevance is calculated so that the higher the number of content words that constitute the topic, the higher the value appears in a narrower range for the sentence and the topic. , For the sentence relevant to the topic, identify the subject by collating with data representing a predetermined statement,
An opinion collecting apparatus characterized in that a document ID, relevance, and a subject are included in the statement data and stored in the storage unit.
In the opinion collecting device according to claim 3,
The computing unit is
An opinion collecting apparatus characterized by referring to the document data for a sentence content for which a statement origin and a statement date and time cannot be specified, and obtaining a default statement subject relating to a document including the sentence for each sentence.
In the opinion collecting device according to claim 4,
The computing unit is
The obtained opinion data is sorted in descending order using the relevance to the topic as the first sort key, and / or when the number of collected document contents exceeds the upper limit value specified as an option An opinion gathering device characterized by discarding the portion where the number of the subsequent ranks exceeds that number.
In the opinion collecting device according to claim 1,
The computing unit is
In the calculation process for calculating the importance level,
With reference to the document data, the past document data of the speaking subject is searched,
The retrieved past document data is divided into sentence units,
For each divided sentence, the sentence is parsed to obtain the structural elements that make up the syntax structure,
For each syntax element of the parsing result, the semantic similarity (Sim) with the above values is calculated,
In the sentence of the syntax element, calculate a context factor representing the influence of the context on the degree of support for values,
From the semantic similarity (Sim) and the context factor, the importance (V) for the value of the sentence is calculated,
An opinion collecting apparatus that calculates a degree of importance of the speaking subject based on the values by accumulating the degree of importance (V) with respect to the values of each sentence.
In the opinion collecting device according to claim 6,
The computing unit is
The main part of the syntax element at the top of the sentence structure of the values and the main part of each syntax element coincide with each other and are registered in predetermined semantic similarity phrase pair data indicating the semantic similarity between phrases. An opinion collecting apparatus characterized in that semantic similarity (Sim) is obtained by examining whether or not it corresponds to a similar phrase pair.
In the opinion collecting device according to claim 6,
The storage unit
For the main part of the syntax element, data of the main factor that defines a formula for calculating its own context factor from the context factor of the lower syntax element;
Auxiliary factor data that converts the influence on the context factor of the auxiliary component corresponding to the sentence end expression of the syntax element into data,
Includes data of modifier part data that quantifies the effect of syntax element modifiers on context factors,

The computing unit is
In the process of calculating the context factor,
Assign the main factor, auxiliary factor, and modifier factor to each syntax element based on the data,
From the designated syntax element to the higher-order syntax element, calculate the main factor of each syntax element according to the calculation formula given by the main factor,
An opinion collecting apparatus characterized in that a context factor is calculated by synergizing an auxiliary factor factor and a modifier factor with a calculated main factor value.
In the opinion collecting device according to claim 6,
The computing unit is
Seeking syntactic structure derived from values,
When the main part of the topmost syntax element of the syntax structure corresponds to a predetermined promotion / suppression word related to the promotion / suppression of an event, the syntax element to be promoted / suppressed is designated as the top-level syntax. Element, and in the case of suppression, set a reverse flag of values,
When the value support (s) of each syntax element is calculated from the semantic similarity (Sim) and the context factor, and an addition or a value inversion flag is set in the importance (V) An opinion collecting device characterized by subtracting.
In the opinion collecting device according to claim 1,
In the display section,
Referring to the importance level data for each subject and value, the importance level for the second value for the vertical axis and the vertical value for the first value for the horizontal axis for each of the message subjects An opinion collecting device, wherein each utterance subject is displayed with ordinate as the ordinate.
In the opinion collecting device according to claim 1,
In the display section,
An opinion collection device characterized in that referring to opinion data, each document content is displayed in association with the display position of the subject of the document content.
In the opinion collecting device according to claim 1,
The remark data further includes an age division,
The computing unit is
Creates importance level data by utterance entity and values for each age group, stores them in the storage unit, and displays on the display unit or outputs to the output unit how the importance level regarding the values of the utterance entity changes An opinion collection device.
An opinion collection system,
A terminal,
An opinion collection device connected to the terminal via a communication network;

The opinion collection device includes:
A plurality of document data including the document content in advance, a plurality of opinion data including the document content and the speaking subject, a storage unit holding importance data for each speaking subject and each value,
An arithmetic unit;
Have

The computing unit is
Receiving a topic entered by the terminal that defines what documents to collect and one or more values that are considered to have an impact when determining the pros and cons of the topic;
Search the document data for the document content related to the received topic,
Finding a set of speaking subjects of the retrieved document content, storing a plurality of opinion data including the document content and the speaking subject in the storage unit,
For each speaking subject included in the opinion data, calculate the importance for each of the values,
Create importance level data for each subject and value from the calculated importance level, store it in the storage unit,
The opinion collection system characterized in that the importance level data for each subject and value is displayed on a display unit or output to an output unit.
In the opinion collection system according to claim 13,
The opinion collection device transmits, to the terminal, the opinion data and the importance level data according to the speaking subject and values,
The terminal
An input unit for inputting the topic and the one or more values;
A terminal storage unit that stores the opinion data received from the opinion collection device, and the importance data according to each speaking subject and each value;
An opinion comprising: a display unit for displaying the opinion data and the importance data for each utterance subject and each value, or an output unit for outputting, and a terminal calculation unit for performing communication, storage, and display processing Collection system.
An opinion collection method in an opinion collection device,
The opinion collection device includes:
A plurality of document data including the document content in advance, a plurality of opinion data including the document content and the speaking subject, a storage unit holding importance data for each speaking subject and each value,
An arithmetic unit;
With

The computing unit is
Receiving a topic entered by the terminal that defines what documents to collect and one or more values that are considered to have an impact when determining the pros and cons of the topic;
Search the document data for the document content related to the received topic,
Finding a set of speaking subjects of the retrieved document content, storing a plurality of opinion data including the document content and the speaking subject in the storage unit,
For each speaking subject included in the opinion data, calculate the importance for each of the values,
Create importance level data for each subject and value from the calculated importance level, store it in the storage unit,
An opinion collecting method, wherein the importance data classified by each speaking subject and each value is displayed on a display unit or output to an output unit.

Images