JP2931553B2 - Topic processing device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a topic processing apparatus, and more particularly, to a topic processing apparatus for activating a conversation between a plurality of participants.

[0002]

2. Description of the Related Art Dialogue can be regarded as a very effective creative act for obtaining new ideas. In academic societies and symposiums, organizers and presenters often say that "we expect lively discussions" because, in part, it is premised that discussions have such an effect. It is also a common experience that researchers can gain new information from the disorganized dialogue they interact with in the lounge, and also gain new ideas and tips for problem solving.

It is said that there are two important processes in the process of human thinking. One is a divergent thinking process, and the other is a convergent thinking process. For a problem whose solution is unknown, a person first gathers not only information that is clearly related to the problem in the divergent thinking process, but also a lot of information that is vaguely related or suddenly anxious. The information fragments collected in this way are integrated in a convergent thinking process. At this time, when a new association is found between information fragments whose apparent association is unknown, a person can obtain a new idea.

[0004]

Some ideas and the like try to obtain ideas more effectively by intentionally separating these two processes intentionally. For example, brainstorming is a method for mainly performing divergent thinking, and the KJ method is a method for effectively performing convergent thinking.

[0005] On the other hand, the two are not usually distinguished in daily conversation. However, from a different point of view, everyday dialogue is thought to be a mixture of these divergent and convergent processes. In other words, various information is provided by the participants' remarks (divergent process). Here, when a participant finds some association between the provided information fragments or between the provided information fragments and some information in his or her knowledge (convergent process), the participant finds the related Presenting new topics based on gender, the dialogue will evolve into a new phase. The dialogue proceeds one after another by repeating the process of divergence and convergence.

[0006] Therefore, a main object of the present invention is to provide a topic processing apparatus which always activates participating dialogues, thereby enhancing the creativity of the dialogues.

[0007]

The invention according to claim 1 is
A topic processing apparatus for activating a dialogue between a plurality of speakers, wherein a morphological analysis is performed using individual utterances of each participant as a utterance object, and only keywords of a predetermined type are extracted as keywords from the result. Extracting means, keyword weight calculating means for evaluating the weight of each keyword included in each utterance object based on the appearance frequency and appearance interval of each extracted keyword, and based on the obtained keywords and their weights, The relevance of all the utterance objects up to a certain point is calculated using the shared relation of the keywords included in each utterance object, while the relevance of all the keywords is co-occurred in each utterance object of those keywords , And as a result, the utterance objects Dialogue structuring means for structuring the dialogue content by expressing the mutual relationships as a spatial structure, and obtaining the spatial position information of the part where speech objects are particularly dense from the obtained dialogue content structure. Main topic detecting means for detecting a main topic in the conversation up to the point in time.

According to a second aspect of the present invention, there is provided a topic processing apparatus for activating a dialogue between a plurality of speakers, including a keyword extracting unit, a keyword weight calculating unit, and a dialog structuring unit. And a topic blank area detecting means for detecting a blank area of a topic in the dialogue up to that time by obtaining position information of a blank area where no comment object exists from the obtained dialog content structure.

According to the third aspect of the present invention, in addition to the respective configurations of the second aspect, the spatial position information of a portion where the utterance objects are particularly dense is obtained from the obtained dialog content structure, so that up to that point in time. And a main topic detecting means for detecting a main topic in the dialogue of the user.

In the invention according to claim 4, in addition to the constitution of claim 1 or 3, a predetermined number of keywords located near the spatial position of the main topic grasped by the main topic detecting means are collected. Main topic-related keyword extraction means for evaluating the weight of the keyword according to the distance from the spatial position of the main topic, associative recall means for recalling the associative keyword group using the associative dictionary based on the obtained keyword group, A main topic related information extracting means for searching and extracting related information from a database using the recalled associative keyword group is included.

According to a fifth aspect of the present invention, in addition to the configuration of the second or third aspect, a predetermined number of keywords located near the topic blank region detected by the topic blank region detecting means are collected. And a blank area surrounding keyword extracting means for evaluating the weight of the keyword according to the distance from the topic blank area, and a blank area related information extracting means for searching and extracting related information from a database using the obtained keyword group.

In the invention according to claim 6, in addition to the structure of claim 3, a predetermined number of keywords located near the spatial position of the main topic grasped by the main topic detecting means are collected, and the weight of these keywords is weighted. Main topic related keyword extraction means for evaluating according to the distance from the spatial position of the main topic, associative recall means for recalling the associative keyword group using the associative dictionary based on the obtained keyword group, and recalled associative keyword A main topic related information extracting means for searching and extracting relevant information from a database using a group, and collecting a predetermined number of keywords located in the vicinity of the identified topic blank area, and weighting these keywords to a distance from the topic blank area Searching related information from the database using the keyword extraction means around the blank area and evaluating the keywords Including a blank area-related information extracting means for output.

[0013]

FIG. 1 is a block diagram showing an embodiment of the present invention. In FIG. 1, a statement is input as text data, and each of the sequentially input texts is referred to as one statement object. This comment object is given to the comment sentence processing module 1 and the topic structuring module 2. The sentence processing module 1 includes a syntax analysis module 11 and a keyword weight calculation module 12, and the syntax analysis module 11 performs a morphological analysis on the input speech object and extracts a keyword group. The keyword group is provided to the topic structuring module 2 and the keyword weight calculation module 12
Given to. The keyword weight calculation module 12 calculates the weight of each keyword and gives it to the topic structuring module 2.

The topic structuring module 2 arranges these in a two-dimensional space, reflecting the association between the keyword object and the comment object obtained up to that point,
Structure the content of the conversation. The two-dimensional spatial arrangement information of the comment object / keyword object is given to the topic development situation recognition module 3. The topic development situation recognition module 3 includes a main topic detection module 31 and a topic blank area detection module 32.
1 detects the position of the main topic, and the topic blank area detection module 32 detects the topic blank position.

FIG. 2 is a flowchart for explaining the operation of the topic structuring module 2 and the topic development situation recognition module 3 shown in FIG.

Next, a specific operation of the embodiment of the present invention will be described with reference to FIGS. The syntax analysis module 11 of the statement processing module 1 parses the input statement object and determines the part of speech of each word. Next, only words for which nouns and parts of speech could not be determined are extracted, and this word group is used as a keyword of the utterance object. Then, the keyword weight calculation module 12 obtains the weight W _wi keyword w _i in the n speech _object, a _n by the following equation.

[0017]

(Equation 1)

[0018] However, if f _wi has appeared in the w _i are a number of remarks object up to the (n-1) th speech object,
f _{wi, n} is either w _i was a number appeared in the n-th speech _object, i _{wi, n} indicates whether w _i has been recently used to before what remarks of the n-th speech object. F _g , F _l , and I are constants, and empirically give values of 5, ₁ , and 10, respectively.

The reason for such weighting is as follows. First, the keywords that appear frequently are either words that are commonly used in any dialogue, or words that relate to the overall topic of the dialogue, and in any case are important in structuring the topic. Low degree.
If such words have a large weight, each topic cluster may be obscured in the topic structuring stage described below.

It is based on the idea that words appearing many times in one utterance have a high importance in the utterance, and that a high-frequency word has a high importance if it is used after being unused for a while.

Next, in step SP1 (abbreviated as SP in FIG. 2) SP1 shown in FIG. 2, the topic structuring module 2 shares the keyword objects in each statement object with each other. The distance between the objects is obtained based on the dual scaling method using the co-occurrence relation. Here, the dual scaling method means that when an object set composed of a plurality of quantification attributes is given, the object set and the attribute set are each given a score quantity, whereby the attribute commonality between objects and the attribute This is a method of expressing the co-occurrence of as a relative positional relationship in space. In step SP2, the top two components having the highest contribution are taken out of the principal components for quantifying the spatial structure, and the position information of each object in the two-dimensional space is determined.

In step SP3, the main topic detection module 31 of the topic development situation recognition module 3 divides the two-dimensional space into 16 × 16 cells. Then, in step SP4, for the n-th cell, the average of the number of utterance objects included in the cell itself and the eight surrounding cells is obtained, and this is set as the weight of the n-th cell. This is 16
Calculate for all × 16 cells. In step SP5, a cell whose weight exceeds a predetermined threshold value θ is selected, this is regarded as a main topic cell, and the coordinates of the center of those cells are output.

On the other hand, in step SP6, the topic blank area detection module 32 applies the Euclidean distance conversion method to the area where a cell having a weight of zero exists, thereby obtaining the center position of the area and the weight of the center ( The distance from the cell area where the weight is not zero, the greater the distance, the greater the weight). The Euclidean distance conversion method is a method of calculating the distance from the center of an area in space and the periphery of the center. As a result, the weight becomes a predetermined threshold φ
Is selected as the center of the topic blank, and the coordinates of the center of those areas are output.

As described above, according to this embodiment, the position of the main topic and the topic blank position can be detected.

FIG. 3 is a block diagram showing another embodiment of the present invention. In FIG. 3, the statement processing module 1
The topic structuring module 2 and the topic development situation recognition module 3 are the same as those in FIG. 1, and in the embodiment of the present invention, a topic providing module starting module (topic conversion detecting module) 4 and a topic providing module 5 are provided. . The topic providing module starting module 4 detects a topic turning point, and the topic providing module 5 extracts information serving as a topic seed according to the situation and provides it to the user at the timing designated by the topic switching detecting module 4. . For this purpose, the topic providing module 5 includes a keyword extraction and weight calculation module 51, an article database search module 52, and an association recall module 53.

FIG. 4 shows the topic providing module 5 shown in FIG.
5 is a flowchart for explaining the operation of FIG. In addition,
The comment processing module 1, topic structuring module 2, and topic development situation recognition module 3 in FIG. 3 perform the same operations as in FIG. 1, detect the main topic position and the topic blank position, and provide them to the topic providing module 5. The topic providing module 5 determines in step SP11 whether or not an activation command has been received from the topic providing module activation module 4. If no activation command has been received, the topic providing module 5 waits. If an activation command has been received, the topic development status recognition module 3 in step SP12. It is determined whether or not a topic blank position is detected from.

If there is a topic blank area, step SP13
At step SP14, the keyword extraction and weight calculation module 51 collects a predetermined number m of keyword objects in order from the one closest to the center of the topic blank area. At the same time, a weight that is inversely proportional to the distance from the center of the topic blank area is obtained for each keyword object. Then, in step SP15, the obtained weighted keyword group is set as a search keyword group. In step SP16, using the search keyword group,
The article database search module 52 performs an article database search and provides search results.

At step SP12, if there is no topic blank area, a main topic cell closest to the recently input utterance object at step SP17 is searched. Then, in step SP18, the keyword extraction and weight calculation module 51 collects a predetermined number m of keyword objects in order from the one closest to the main topic cell center. At the same time, a weight that is inversely proportional to the distance from the center of the main topic cell is obtained for each keyword object. In step SP19, the associative recall module 53 uses the obtained weighted keyword group to perform associative recall by an associative event that the system has in advance, and obtains an associative keyword group. Then, in step SP20, the acquired associative keywords are used as search keywords, and in step SP16, the article database search of the article database search module 52 is performed using the search keywords to provide search results.

[0029]

As described above, according to the present invention, morphological analysis is performed using individual utterances of each participant as utterance objects, and only words of a predetermined type are extracted as keywords from the results, and each extracted word is extracted. Evaluate the weight of each keyword included in each statement object based on the appearance frequency and appearance interval of the keyword, and based on the obtained keywords and their weights, relevance of all statement objects to a certain point in time Is calculated using the shared relationship of the keywords included in each statement object, while the relevance of all the keywords is calculated using the co-occurrence relation of each keyword in each statement object, and the statement objects and keywords are calculated. Structuring dialogue by expressing mutual relationships as a spatial structure The main topic in the dialogue up to that point is detected by obtaining spatial position information of the part where the speech objects are particularly dense from the dialogue content structure, so that topic acquisition processing can be performed for any type of dialogue. Can be obtained, and the development direction of a new topic can be found by detecting a blank area. In addition, it is possible to acquire topics not only in the purpose thinking type but also in the free expansion type dialogue, and it is not necessary to set the direction of the topic to be newly developed as a plan in advance,
According to the situation of the dialogue, it is possible to acquire a new development direction from the situation itself.

[Brief description of the drawings]

FIG. 1 is a schematic block diagram of an embodiment of the present invention.

FIG. 2 is a flowchart for explaining operations of a topic structuring module and a topic development situation recognition module shown in FIG. 1;

FIG. 3 is a block diagram showing another embodiment of the present invention.

FIG. 4 is a flowchart for explaining the operation of the topic providing module shown in FIG. 3;

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Comment sentence processing module 2 Topic structuring module 3 Topic development situation recognition module 4 Topic providing module starting module 5 Topic providing module 11 Parsing module 12 Keyword weight calculation module 31 Main topic detection module 32 Topic blank area detection module 51 Keyword extraction and Weight calculation module 52 Article database search module 53 Associative recall module

Continuation of the front page (56) References JP-A-6-236410 (JP, A) JP-A-5-81328 (JP, A) JP-A-7-230468 (JP, A) Yasuyuki Kado, Kenji Mase, "Group Visualization of Topic Space in Discussion ”, IEICE Technical Report, Vol. 95, No. 237 (HCS95-19), p. 83-p. 90 (1995) (58) Field surveyed (Int. Cl. ⁶ , DB name) G06F 17/20-17/30 JICST file (JOIS)

Claims (6)

(57) [Claims] 1. A topic processing apparatus for activating a conversation between a plurality of speakers, performing a morphological analysis using individual utterances of each participant as a utterance object, and extracting only words of a predetermined type from the results. Keyword extraction means for extracting as keywords, keyword weight calculation means for evaluating the weight of each keyword included in each utterance object based on the appearance frequency and appearance interval of each keyword extracted by the keyword extraction means; Means and the weights obtained by the keyword weight calculation means and based on the weights thereof, calculate the relevance of all the utterance objects up to a certain point in time using the sharing relationship of the keyword group included in each utterance object, All keywords are related to each other Dialog structuring means for calculating the dialog content by using the co-occurrence relationship of each of the utterance objects of the keywords and expressing the relationship between the utterance objects and the keywords as a spatial structure, and the dialog structuring A topic processing device comprising a main topic detection unit for detecting a main topic in a conversation up to that point by obtaining spatial position information of a part where speech objects are particularly dense from a conversation content structure obtained by the means. 2. A topic processing apparatus for activating a conversation between a plurality of speakers, performing a morphological analysis using individual utterances of each participant as a utterance object, and extracting only words of a predetermined type from the result. Keyword extraction means for extracting as keywords, keyword weight calculation means for evaluating the weight of each keyword included in each utterance object based on the appearance frequency and appearance interval of each keyword extracted by the keyword extraction means; Means and the weights obtained by the keyword weight calculation means and based on the weights thereof, calculate the relevance of all the utterance objects up to a certain point in time using the sharing relationship of the keyword group included in each utterance object, All keywords are related to each other Dialog structuring means for calculating the dialog content by using the co-occurrence relationship of each of the utterance objects of the keywords and expressing the relation between the utterance objects and the keywords as a spatial structure, and the dialog structuring A topic processing apparatus comprising: topic blank area detecting means for detecting a blank area of a topic in a dialogue up to that time by obtaining position information of a blank area where no comment object exists from a dialog content structure obtained by the means. 3. A main topic for detecting a main topic in a dialogue up to that point by obtaining spatial position information of a part where speech objects are particularly dense from the dialogue content structure obtained by the dialogue structuring means. 3. The topic processing device according to claim 2, further comprising topic detection means. 4. A method for collecting a predetermined number of keywords located near the spatial position of the main topic grasped by the main topic detecting means, and evaluating the weights of these keywords according to the distance from the spatial position of the main topic. A main topic-related keyword extracting unit, an associative recalling unit that recalls an associative keyword group using an associative dictionary based on the keyword group obtained by the main topic-related keyword extracting unit, and an associative keyword group recalled by the associative recalling unit 4. The topic processing apparatus according to claim 1, further comprising a main topic related information extracting means for searching for and extracting related information from a database by using a keyword. 5. A blank area which collects a predetermined number of keywords located near the topic blank area detected by the topic blank area detecting means, and evaluates the weight of these keywords according to the distance from the topic blank area. 4. The topic processing according to claim 2, further comprising a peripheral keyword extracting unit, and a blank region related information extracting unit that retrieves and extracts related information from a database using the keyword group obtained by the blank region surrounding keyword extracting unit. apparatus. 6. A method according to claim 1, further comprising the step of collecting a predetermined number of keywords located near the spatial position of the main topic detected by the main topic detecting means, and evaluating the weights of these keywords in accordance with the distance from the spatial position of the main topic. A main topic-related keyword extracting unit, an associative recalling unit that recalls an associative keyword group using an associative dictionary based on the keyword group obtained by the main topic-related keyword extracting unit, and an associative keyword group recalled by the associative recalling unit. Main topic related information extracting means for searching for and extracting relevant information from a database using a predetermined number of keywords located in the vicinity of the topic blank area grasped by the topic blank area detecting means; Means for extracting keywords around blank areas to be evaluated according to the distance from 4. The topic processing apparatus according to claim 3, further comprising a blank area related information extracting unit that searches and extracts related information from a database using the keyword group obtained by the blank area surrounding keyword extracting unit.