JPH04186435A - Interaction control method and interaction control device - Google Patents

Abstract

PURPOSE:To execute interaction management capable of corresponding to interactions mixedly including a system initiative type interaction and a user initiative type interaction by means of simple processing by adopting constitution for controlling an input means and a response means by communication through a common memory and constitution for storing a solution or the like detected in the response means by a user in the common memory. CONSTITUTION:When the user inputs a question from a terminal 1, the question is sent to an input part 41, its natural language analyzing processing is executed and the analyzed result is stored in a working memory 5 as a user's input. An interaction control part 43 executes context processing such as the grasp of abbreviation and stores the processed result in the memory 5 as a user's inquiry and a retrieving part 44 retrieves the solution of the inquiry and stores the retrieved result in the memory 5. A response part 42 responds to the user by the display or the like of the retrieved result through the terminal 1, and when the user applies any operation to the retrieved result, stores the operated contents in the memory 5. Consequently interaction management capable of corresponding to an interaction mixedly including a system initiative type interaction and a user initiative type interaction can be executed by simple processing.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a user interface between a machine and a human, and particularly relates to an interaction control method and apparatus that can flexibly respond to human interaction.

[Prior Art] Dialogue in a user interface such as a conventional expert system is basically controlled in a system-driven manner. A system-driven interface is a guided interface in which the system unilaterally makes inquiries and responses, and the user answers them. For example, when responding to a user's question, in the conventional dialog method,
As described in Japanese Unexamined Patent Publication No. 106042/1983, the method of responding to the user was changed depending on the number of solutions found. For example, if there are a large number of solutions, only the number of searched solutions will be responded to, and if there are few, all search results will be responded to.

[Problem to be solved by the invention]

As mentioned above, all interactions in conventional user interfaces are system-driven. This is not a problem for users who have little experience using the system, but users who have a certain amount of experience find this type of dialogue annoying. One response method is to use a multi-window to display all solutions in card format even when there are many search results. This allows the user to freely examine the search results. Here, the act of the user considering the search results also becomes important information for dialogue. However, in the past, this response function was only for displaying information, so it was not possible to control the dialogue that reflected the user's actions. That is, a flexible user-driven user interface that allows the user to freely examine search results and make adjustments to the system based on the results of the examination has not been realized.

An object of the present invention is to realize dialogue management capable of handling a mixture of system-driven and user-driven dialogues through simple processing.

[Means to solve the problem]

In order to achieve the above purpose, questions from users are input, analyzed by an input means, the results are stored in a common memory as user input, and user questions are generated from the above user input by an interaction control means. and stored in the common memory, a search means searches for an answer to the user question, the result is stored in the common memory, and the response means outputs the search result to the user.

Further, a response from the user to the search result is stored in the common memory by the response means.

The response means outputs search results for the user's questions,
Among the search results, the method includes a step of storing solutions viewed by the user on a common memory.

For example, a step is provided to display the search results of a user's question in the form of a card, and store the solution displayed at the top of the card in a common memory each time the user turns over the card.

Further, when storing the solution on the common memory, a step is provided in which the solution is stored only when the solution is displayed for a certain period of time or more.

[Effect]

By adopting a configuration in which the input means and the response means are controlled by communication via a common memory, and a configuration in which the response means stores the solutions seen by the user on the common memory, the user can display them in the form of a card, etc. The understanding of the act of checking search results can be realized independently of normal question/answer type dialog processing, and dialog control can be simplified. This allows users to enter new questions after the search results are displayed.
You can also freely examine search results displayed in card format, regardless of order.

Furthermore, by storing the solution seen by the user on the common memory, the user model of the user is expressed on the common memory, and this information can be used by both the input analysis means and the response means. In addition, when storing solutions in the common memory, the solution is stored only when the solution has been displayed for a certain period of time, so if the user is looking at the search results in pieces at high speed, it is possible that the user is not looking at the solution. I can judge.

〔Example〕

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

FIG. 1 is a diagram showing a conceptual configuration of an example in which the present invention is implemented in a user interface of an expert system that provides guidance on recreational areas using an inference system based on a blackboard model.

In FIG. 1, when a user inputs a question (hereinafter referred to as a question) from the terminal 1, the question is sent to the input unit 41, where it is subjected to natural language analysis processing such as morphological analysis, syntactic analysis, and semantic analysis. , the results are stored in the common memory 5 (hereinafter referred to as working memory) as user input. When the user input, which is the input analysis result, is stored in the working memory 5, the dialogue control unit 43 performs context processing such as supplementing omissions, and stores this in the working memory 5 as the user's adjustment. . When the user's arrangement is stored in the working memory 5, the search unit 44 searches for a solution for the arrangement,
The search results are stored in the working memory 5. Thereby, the response unit 42 responds by displaying the search results to the user through the terminal 1. Furthermore, if the user performs any operation on the search results displayed on the terminal 1, the response unit 42 stores the contents in the working memory 5. In this way, each processing section such as the input section 41 and the response section 42 performs processing while communicating via the working memory 5. This type of inference (control) method is called inference using a blackboard model.

Here, first, the inference system using the blackboard model will be specifically explained below.

In FIG. 2, 1 manages an inference program 31 and a working memory 5, which will be described later, for accepting input from the user and executing various inferences between the user and the system based on questions and responses from the system. Program 32 for
A program 33 for importing character strings input from the terminal 1 and displaying search results on the terminal 1, a program 34 for natural language analysis such as morphological analysis, syntactic analysis, and semantic analysis, and a program 34 for analyzing map information and recreational areas. This is a program memory for storing programs 35 and the like for executing various search processes for the database 7. 5 is a working memory used as a common memory; 6 is a conversation stack used for generating user questions; 7 is a database for storing various data; 8 is a system for connecting each means;
This is a bus for making phone calls.

The knowledge for making inferences is IF- as shown in Figure 4.
It is expressed as a THEN type production rule and stored in the knowledge base memory 4. Each stored rule is selected/executed according to the execution result using the program in the program memory 3 and the content of the data in the working memory 5. here.

Data to the working memory 5 is stored, for example, in the format shown in FIG. These are a class name of data, an ID name of data, and a set of attributes and attribute values of the data. For example, if the person is 30 years old and male, [class name = human, id master human 1 attribute - age = 30, attribute -
Gender = male.

Now, in the production rule of FIG. 4, the data of [class name = input] is not stored in the working memory 5, and the program 33 that imports the character string input from the terminal 1 in the program memory 3 can be executed. (i.e., when the user inputs a string), the execution part (TH
EN part) can be executed. The execution part of the figure is [
This is an instruction to store the data of class name=input, attribute-text=input character string in the working memory 5. In the execution part,
In addition, instructions for deleting data already stored in the working memory 5 and instructions for changing attribute values of stored data can also be written. The execution of the instructions is actually performed by the program 32 in the program memory 3.

Now, actual inference is realized by the inference program 31 stored in the program memory 3.

The flow of the process is shown in FIG. This program first searches all rules stored in the knowledge base memory 4 shown in FIG. 2 for all rules whose condition parts are executable. The rule named Writer's Power in Figure 4 is in the working memo υ5 [class name = data in the box is not stored.

This is selected when the program 33 that captures a character string input from the terminal 1 in the program memory 3 can be executed. Here, if there is no searched rule, the inference ends, but if there is one or more, one of the searched rules is selected. For example, select in the order in which they are stored starting from the beginning.

When one rule is selected, the execution part of the selected rule is executed. In the case of the literati rule shown in FIG. 4, the data of Loclas name=input is stored in the working memory 5. Inference is made by repeating the above process. Here, the rule in FIG. 4 is such that the condition part cannot be executed due to the change in the working memory 5.
Then it will no longer be selected. In this way, since each rule operates data-driven according to the contents of the working memory 5, flexible control of knowledge is realized.

In FIG. 6, it is inefficient to always evaluate the condition parts of all rules stored in the knowledge base memory 4. One way to solve this problem is the concept of knowledge sources. A knowledge source consists of multiple rules, and like rules, it has a condition part for activation. Then, only when the condition is executable, the rule included in the knowledge source is executed as shown in FIG. Then, when there are no rules that can be evaluated among the rules included in that knowledge source, lx
Execution of J knowledge is terminated. In other words, inference program 3
1 first searches for an operational knowledge source and then executes the rules of the searched knowledge source. By repeating this process, inference can be made efficiently.

Next, using the above blackboard model inference system,
An example in which the present invention is implemented in a user interface of an expert system that provides guidance on recreational areas will be described.

In this embodiment, as explained in FIG. 1, four knowledge sources as shown in FIG. 7 are provided. As shown in FIG. 8, the input knowledge source 41 (corresponding to the input section in FIG. 1) includes rules for importing input from the terminal l as shown in FIG. Consisting of It is activated when there is an input from the user, and writes the analysis result of that input (user input in FIG. 10) into the working memory 5.

The search knowledge source (corresponding to the search section in FIG. 1) 44 is.

It is an expert system that provides guidance on recreational areas, and search requests (user timing in Figure 10) are stored in working memory 5.
It is activated when the data is written to , performs the search, and writes the result (system response in FIG. 10) to the working memory 5. There is also a rule that, for example, when there are too many solutions searched, an inquiry to the user to reduce the number of solutions (system adjustment in FIG. 10) is generated and written into the working memory 5. A knowledge source for response (corresponding to the response section in FIG. 1) 42 is a knowledge source for search (corresponding to the response section in FIG. 1) that conveys search results and inquiries from the search section 44 to the user through the terminal 1.A knowledge source for dialog control. The input knowledge source (corresponding to the dialogue control unit in FIG. 1) 43 understands the input content from the input analysis result and context information generated by the input knowledge source (input unit) 41, and controls each other knowledge source accordingly. This can be realized, for example, by the method described in Japanese Unexamined Patent Publication No. 106042/1983.

A dialog control method according to an embodiment of the present invention will be explained using the conversation example shown in FIG. First, the user from terminal 1 can ask which resort in Tokyo has the ULR pool shown in Figure 11? ”, the input knowledge source (input unit) 41 is started and the analysis result of the input is written into the working memory 5. This corresponds to the user input in FIG. When the user input is written into the working memory 5, the dialog control knowledge source (dialogue control unit) 43 is activated, performs context processing, and then writes the result to the stack 6.

As shown in FIG. 12, input 1 is the first input, so nothing is stored in the stack 6. Therefore, the input is pushed onto the stack 6 without performing any context processing, and the user input in the working memory 5 is turned into a user question. As a result, the search knowledge source (search unit) 44 is activated.

The search knowledge source (search unit) 44 is a user question (ul=r
Which resort in Tokyo has a pool? ”) and write the solution into the working memory 5. This solution is conveyed to the user by the response knowledge source (response unit) 42. For example, as shown in FIG. 13, the information is output in a card format. In other words, the user can see the search results on the display as if flipping over a card. The response knowledge source (response unit) 42 registers the information on the resort displayed at the top of the card on the blackboard as shown in FIG. 14 every time the user turns over the search results displayed in card format. . Furthermore, the name of the registered recreational area is recorded and added to the known attributes of the user question (FIG. 10).

Next, when the user asks, "Where is the manual road?", the dialog control knowledge source (dialogue control unit) 43 examines the relationship with the input already stored in the stack 6.

Stack 6 stores u 1 = "Where is the resort in Tokyo with a swimming pool?", but there is no relationship between this sentence and the current input "Where is the hand road?" Therefore, the input is sent to the stack 6 and the user input in the working memory 5 is made into a user question. As a result, the search knowledge source (search unit) 44 is started, the hand road is searched, and the response knowledge source (response unit) 42 starts the 15th
It will be displayed as shown.

Suppose that the user then asks, ``Where is the cheapest place along the road?''

The dialogue control knowledge source (dialogue control unit) 43 examines the relationship with inputs already stored in the stack 6.

Here, the input is the question (
Which recreational area in Tokyo has a ul=r pool? ”).

In other words, which places are cheaper than the tourist resorts that users know about along R-teyo Kaido? ” can be interpreted as meaning. Here, what is meant by "a recreational area known to users"?
This is the recreational area that the user saw as the previous response, that is, the recreational area that was registered on the blackboard by the response knowledge source (response unit) 42.

In this way, by recording the dialogue history on the blackboard, the dialogue control knowledge source (response unit) 43 searches the blackboard for user questions related to the inputs stored in the stack 6, and By simply reading the attribute values, it is possible to obtain the ``relaxation destinations that the user knows''. That is, it is possible to easily implement user-initiated dialog context processing that reflects the dialog situation.

Additionally, if there are many search results for user question u1, the system displays the search results in card format, and also displays the message ``How much is your budget?'' ”. Thereby, the user may respond to this arrangement, or it is also possible to make a user-initiated arrangement after considering the search results displayed in the card format as described above.

〔Effect of the invention〕

According to the present invention, system-driven interaction in which the system efficiently guides the search, and user interaction in which the user freely examines the search results and makes adjustments to the system based on the results of the examination are possible. Dialogue management that can handle a mixture of initiative-driven dialogues can be achieved with simple processing.

[Brief explanation of the drawing]

Figure 1 is a conceptual configuration diagram of an embodiment of the present invention, Figure 2 is a block diagram of an embodiment of the present invention, Figure 3 is a detailed diagram of the contents of the program memory, and Figure 4 is for knowledge base. An example of a production rule stored in memory, FIG. 5 is an example of a data structure written to a working memory, FIG. 6 is a flowchart of an inference program, and FIG. 7 is an example of knowledge in the knowledge base memory of an embodiment of the present invention. Figure 8 is a diagram showing the configuration of each knowledge source in the embodiment, Figures 9 and 10 are examples of data structures written to the working memory in the embodiment,
Figure 11 is an example of a conversation, Figure 12 is a diagram showing the conversation stack management process, Figure 13 is a display example of the response section, and Figure 14 is the search results stored in the working memory and viewed by the user. FIG. 15 is a diagram illustrating an example of the display of the response section. Explanation of symbols 1...terminal, 5...working memory (common memo January,
41... Knowledge source for input (input unit), 42... Knowledge source for response (response unit), 43... Knowledge source for dialog control (dialogue control unit), 44... Knowledge source for search (search unit).

Claims (1)

[Scope of Claims] 1. Inputting a question from a user, analyzing it with an input means, and storing the result in a common memory as a user input, and transmitting the user question from the user input to the dialogue control means. search for the solution to the user question using the search means and store the result in the common memory; output the search result to the user using the response means; An interaction control method characterized in that a response from a user to a search result is stored in the common memory by the response means. 2. The dialog control method according to claim 1, wherein the dialog control means performs context processing on the user input to generate a user question. 3. The dialog control method according to claim 1, wherein the search means generates a system alignment and stores it in the common memory. 4. The dialog control method according to claim 1, wherein the response means displays the search results in a card format. 5. The dialog control method according to claim 1, wherein, as a response from the user to the search result, the solution seen by the user is stored in the common memory by the response means. 6. The interaction control method according to claim 5, wherein when the solution viewed by the user is stored in the common memory by the response means, the solution is stored only when the solution is displayed for a certain period of time or more. 7. The interaction control method according to claim 1, wherein the response means displays the search results in the form of a card, and each time the user turns over the card, the solution displayed at the top of the card is stored in the common memory. 8. In a dialogue control device that is equipped with a common memory and performs dialogue control using the common memory, an input that takes a question from a user as input, analyzes the input, and stores the result as user input in the common memory. means for generating a user question from the user input and storing it in the common memory; searching means for searching for a solution to the user question and storing it in the common memory; a response means for outputting to a user and storing a response from the user to the search result in the common memory; names of the common bus; the input means; the dialogue control means; the search means; and the response means. A dialogue control device comprising: means for connecting and communicating with each other;