JPH0748219B2 - Conversation control system - Google Patents

Description

The present invention relates to a conversation control method for realizing a flexible man-machine interface in a conversation type information processing system such as an information retrieval system.

[Prior Art] A conventional method for advancing conversation between a person and a computer is disclosed in
Like -106042, because it is a method of controlling conversation by referring to a topic sequence management table that has been determined in advance, the flow of conversation is system-driven and uniform, and flexible and natural conversation depending on the situation. There is a problem that cannot be realized. Further, as in Japanese Patent Laid-Open No. 63-153631, a user-initiated conversation control method that attempts to realize a smooth conversation by managing the topic by regarding the search result of the search request from the user as the user's attention point. However, since the response of the system depends on the information retrieval result, the conversational information processing system that supports the information retrieval in the ambiguous retrieval method and provides the user with intellectual assistance. Is not suitable as a conversation control method.

[Problems to be Solved by the Invention]

As described above, in the conventional technology, the conversation control method of the conversation-type information processing system is uniform such as system-driven type or user-driven type. There was a problem that it was difficult to realize flexible conversation control corresponding to.

An object of the present invention is to provide a flexible conversation control system in which system-led conversation and user-led conversation are coordinated.

[Means for Solving the Problems]

In order to achieve the above-mentioned object, an input processing unit that interprets a search request input by a user and a response processing unit that guides and responds from the system together with the input processing unit can operate cooperatively. This is a system in which conversation control is performed by estimating topic transitions, expressing them as one topic tree, and using topic processing means for unified topic management.

[Action]

The input processing means interprets the search request from the user and operates to control the user-led conversation. Also,
The response processing unit refers to the topic tree that reflects the conversation status, performs appropriate guidance and response from the system, and operates to control the system-driven conversation. Further, the topic processing means estimates the transition of the topic, and centrally manages the topic tree with good consistency, so that the topic processing means operates to coordinate the user-led conversation and the system-led conversation. With such a configuration, it is possible to realize flexible man-machine conversation control according to the conversation state.

〔Example〕

 An embodiment of the present invention will be described below with reference to FIG.

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

The search request input by the user is subjected to syntax / semantic analysis in the input processing unit 1 and is interpreted as one keyword representing a search request or a set of a plurality of keywords. The analysis dictionary 12 records a dictionary required for syntax / semantic analysis. This includes a morpheme dictionary, a syntactic dictionary, a semantic dictionary, etc., if necessary, but they are not shown. The input interpretation unit 11 refers to the analysis dictionary 12 to search requests from the user for morphological analysis, syntactic analysis,
Semantic analysis is performed, and the result is sent to the topic processing unit 3 as a set of keywords.

The input interpretation unit 11 can predict the search request from the user by referring to the conversation state estimated and recorded by the topic processing unit 3, and interpret the search request with higher accuracy. You can

In this way, the input processing unit 1 is a part that executes a user-initiated conversation function, which mainly focuses on the interpretation of the search request from the user.

On the other hand, the response processing unit 2 refers to the conversation state estimated and recorded by the topic processing unit 3 to make a response from the system to the user.

The response sentence dictionary 22 records a response sentence template according to various situations, a syntax rule for generating a response sentence, and the like. The response sentence generator 21 refers to the response sentence dictionary 22 to generate a guide or response from the system according to the conversation state, and presents it to the user. For example, when some candidates are obtained as search results, the contents of these candidates are presented to prompt the user to select or narrow down the candidates, or the information of the search request is insufficient. If it is ambiguous, a response sentence requesting supplementation or confirmation of information from the user is presented.

As described above, the response processing unit 2 is a unit that executes a system-initiated conversation function, which mainly includes a guide and a response from the system.

The topic processing unit 3 stores a plan recording unit 32 for recording a basic plan for estimating the transition of a topic associated with the progress of a conversation, and the estimation result of the transition of the topic up to that time in a tree structure (or It has a topic recording unit 33 for recording a topic tree expressed as a combination of words having a network structure), and the topic management unit 31 inputs a search request from a user and outputs a response from the system. Estimate the transition of accompanying topics and update the topic tree.

The processing in the topic management unit 31 has the following contents.

One is that the keyword (or a combination thereof) representing the contents of the search request interpreted by the input processing unit 1 is recorded in the topic recording unit 33 with reference to the basic plan recorded in the plan recording unit 32. This is a process of estimating a transition of a topic from the current topic tree, adding a keyword to a position corresponding to the topic tree to create a new topic tree, and recording the new topic tree in the topic recording unit 33. This is a process for reflecting the state of user-led conversation progress in the topic tree.

The other is a process of estimating the transition of the conversation state according to the system response generated by the response processing unit 2 and presented to the user, and adding the content thereof to the topic tree recorded in the topic recording unit 33. The topic tree is updated according to the system response to the user such as display of a menu of search candidates, request for input of missing information (addition of search conditions, etc.), confirmation message for ambiguous input information, and the like. This is a process for reflecting the state of system-driven conversation progress in the topic tree.

In this way, the topic management unit 31 performs the two types of conversation state estimation processing, but since the estimation results are reflected in one topic tree recorded in the topic management unit 33, the topic tree It is possible to realize management uniformly and with good consistency. That is, it becomes possible to perform conversation control in which a user-led conversation and a system-led conversation are coordinated, and a flexible interface can be realized.

In addition, the information search unit 4 refers to the topic tree recorded in the topic recording unit 33 to create a search condition expression for information search, searches a database (not shown), and obtains the obtained search results. By reflecting in the topic tree, the conversation between the system and the user can be conducted more smoothly.

As is apparent from the above description, it is not necessary for the input processing unit 1 and the response processing unit 2 to further proceed to the information search unit 4 in synchronization with each other. Each processing unit is asynchronous
Even in a processing environment that operates independently, the topic processing unit 1
Since the topic trees are centrally managed, the consistency in the man-machine interface is not disturbed.

An embodiment of the present invention will be described more specifically by taking a conversation between a user and a system in a resort search system as an example.

FIG. 2 shows an example of the basic plan recorded in the plan recording unit 32 when a playground guide is assumed.

FIG. 3 shows how the conversation progresses according to the system guide, along with the corresponding change in the topic tree. (In the topic tree, the part enclosed in [] indicates that the corresponding information has been presented to the user by the system, and the part enclosed in a rectangle indicates that the corresponding information has been input by the user. The same applies to the following figures).

In the basic plan of the vacation guide, the choice of sports or driving is first assumed, and the topic processing unit 3 creates a topic tree based on this. On the other hand, the response processing unit 2
Then, guide to the user, such as S11 "I'll be a tourist guide. Do you enjoy sports or drive?" In response to this guide, the topic processing unit 3 changes each of the excursions, sports, and drives in the topic tree to [excursion], [sports], and [drive], and presents the corresponding information to the user. Update to indicate that it is in the completed state. This topic tree indicates that the topic is predicted to shift from leisure to sports to driving. On the other hand, the input from the user U11 "I want to drive" is interpreted by the input processing unit 1, and the keyword "drive" is passed to the topic processing unit 3. At this time, it is apparent that the input processing unit 1 can interpret the input with higher accuracy by predicting the input from the user with reference to the topic tree. The topic processing unit 3 updates the topic tree to indicate that the user has selected a drive. This reflects a situation where it seems that the topic has shifted to driving. Then, by referring to the basic plan, it can be understood that the topic may change depending on whether the driving place is the city center or the suburbs. Therefore, as shown in FIG. Update.

In the conversation example of FIG. 3, the system leads the conversation according to the basic plan of the system. This is a case seen when the user is not aware of a clear search procedure in advance, and it can be said that the information search system is used as a kind of decision support system.
In such a case, it can be said that the information is vaguely narrowed down by referring to the guides and responses from the system, but there are cases in which the user proactively presents the search conditions with some system response as a hint. Not a few. FIG. 4 shows an example of such a conversation.

In Fig. 4, at the stage where the system-driven type is narrowing down even though it is ambiguous to "whether it be the sea or the mountain as a drive to the suburbs", for example, the user associates the word "sea" with swimming. U21 shows how to update the topic tree, using the search condition "pool is good" as an example.
As indicated by the topic tree in the figure, the search condition of the pool is not consistent with the narrowing down (whether it is the sea or the mountain) up to that point, so the topic transition is re-estimated to update the topic tree.

Further, in FIG. 4, while the guide for the options of S22 "Public pool and hotel pool" was being guided, the relevant information search was completed in the information search section 4, and the P, Q, R pools could be searched. An example response for the case is also shown.

FIG. 5 shows a conversation example in the case where the search condition is input by the user. In this case, since the user's wishes and search plan are clear, the system assists the user in efficiently searching for information. In the figure, if there are many candidates in the search condition only for campgrounds and it is not possible to narrow down the search conditions, we will show how the topic tree changes when a guide to add search conditions such as S32 "Sea or mountain" is given. is there.

When a user-led conversation proceeds, it is important to proceed when the user's search procedure (whether the user is aware or not) does not match the basic search plan of the system. Yes, FIG. 6 shows how the topic tree is updated in such a case. U41 The search condition from the user "I want to go fishing" cannot be estimated from the basic plan (Fig. 2) recorded in the plan recording unit 32, so the topic tree contains "fishing". It is shown only that the key word is input. In the response processing unit 2, since the topic tree does not show specific information for generating an appropriate guide for the user, a very general guide such as S42 "What kind of place is good?" Do. In response to this, the topic processing unit 3 updates the topic tree so as to indicate that the topic can transit to “place”. Then U4
2 If the user adds a search condition that "the sea is good", the topic processing unit 3 checks the basic plan in the system, updates the topic tree as shown in Fig. 6, and conducts a conversation. to continue.

At this time, if the information search unit 4 can search for some “sea as a resort”, it is determined from the topic processing unit 3 to the information search unit 4 whether or not there is information regarding the possibility of “fishing” among these candidates. By re-inquiring, you can search for a resort that you think you want, update the topic tree, and continue the conversation.

Also, if the search cannot be narrowed down only by "the sea as a resort", the topic tree is not updated, and in the response processing unit 3 that refers to the topic tree, "Do you want to swim?" A question is generated, and the user is prompted to enter further search conditions to proceed with the conversation.

The topic tree that is updated as the conversation between the user and the system progresses reflects the search plan unique to the user. Therefore, when the information search is completed, the customized topic tree recorded in the topic recording section 33 is additionally recorded in the plan recording section 32, and the basic plan in the system is increased to make it more flexible. It becomes possible to realize a good conversation control.

〔The invention's effect〕

According to the present invention, as described above, it is possible to provide a flexible man-machine interface according to the conversation state between the user and the system, and to realize an easy-to-use conversational information processing system. The effect of contribution is extremely large.

[Brief description of drawings]

FIG. 1 is a configuration diagram of an embodiment of the present invention, FIG. 2 is a diagram showing an example of a basic plan when a resort search is assumed, FIG. 3 is a diagram showing a system-initiated conversation example, and FIG. FIG. 5 is a diagram showing an example of a conversation in which a system initiative is transferred to a user initiative. FIG. 5 is a diagram showing an example of a conversation in which user initiative and system initiative are mixed.
The figure is a diagram showing an example of conversation when it is difficult to estimate a topic in the system. 1 is an input processing unit, 11 is an input interpretation unit, 12 is an analysis dictionary, 2
Is a response processing unit, 21 is a response generation unit, 22 is a response sentence dictionary, 3 is a topic processing unit, 31 is a topic management unit, 32 is a plan recording unit, 33 is a topic recording unit, and 4 is an information retrieval unit.

Claims (1)

[Claims] 1. An input processing means for analyzing an input search request sentence to extract a keyword, and a topic processing means for selectively outputting a part of a plurality of topics stored in advance for the keyword. And a response processing unit that generates and outputs a part of the topics as a response sentence, and waits for the input of the next search request sentence to the response sentence, wherein the topic processing unit is Regarding topics, a basic plan storage means for storing in advance a basic plan in which a plurality of topics are hierarchically combined, and a part of the topics output from the response processing means are extracted from the next search request sentence. If there is a topic that matches the key word, a plurality of topics hierarchically linked to the matching topic are output to the response processing means as the partial topics, and the response processing procedure is output. If there is no topic that matches the keyword extracted from the next search request statement among the above-mentioned partial topics output from the above, the next search request statement from the topics stored as the basic plan A topic matching unit that detects a topic that matches the keyword extracted from the topic and outputs a plurality of topics hierarchically linked to the matching topic to the response processing unit as the partial topic. Conversation control system.