JPH01158537A - Inference system - Google Patents

Abstract

PURPOSE:To realize an inference system to be able to correct the error of an interactive input with a standard mechanism independent of the object knowledge and a method with high responsibility of the interactive by restoring the inference condition to the condition before updating and returning the inference up to the requested interactive point of time when the instruction of a user is executed in an interactive type knowledge base system. CONSTITUTION:In a knowledge base system having the interactive means with a user, a means 105 to store the history of the updating for the inference condition stored into a inference condition storing part 103 and a means 107 to refer to the stored updating history, restore the inference condition stored into the inference condition storing part 103 to the condition before udpating and return the inference are provided. The history of the updating operation and the history of the interactive with a user for the inference condition for the cycle of the inference are stored, and when the instruction of the user is executed, these history are traced in the sequence of the new one, the inference condition is restored in the condition before updating and the inference is returned up to the requested interactive point of time. Thus, the inference system to be able to correct the error of the interactive input information can be realized with the standard mechanism independent of the object knowledge and the method with the high responsibility of the interactive.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to an inference method in a knowledge-based system,
In particular, this invention relates to an inference method suitable for carrying out processing through dialogue with users, such as in consultation services.

[Conventional technology]

When putting an interactive system into practical use, it is important to be able to correct information that the user has entered incorrectly, as a requirement for improving usability.

In a knowledge base system that has a configuration in which one target knowledge and its inference means are separated and independent, there are two main positions on whether or not to incorporate a correction function that meets the above requirements into the standard inference mechanism.

Most of the knowledge base system construction support tools that have been commercialized so far take the latter position, and do not have the above-mentioned correction function, so it is necessary to deal with each application system individually.

An example of an attempt to take the former position is the Information Processing Society of Japan's 33rd (late 1986) National Conference Proceedings, 4M-11.
(1986), pp. 1367-1368. In this report, an attribute indicating whether or not resetting is allowed is given to the value that is the premise of the inference and the value obtained as a result of the inference.
- Even after inference has been completed, if the user changes the value as necessary, it has been shown that this can be handled by redoing the inference while referring to the attributes of the value.

[Problem that the invention seeks to solve]

Among the conventional technologies mentioned above, construction support tools that do not include the correction function in the standard inference mechanism have the burden of development to implement the correction function individually for each application system, and lack of problem-solving knowledge in the target field. Since the description of the process for realizing the correction function is mixed, there are problems such as impairing the maintainability and expandability of the knowledge.

On the other hand, the technology reported above, which takes the position of including the correction function in the standard reasoning mechanism, has a problem in that it does not take into account the impact of the effort required for correction processing on the responsiveness of dialogue. Rather, this conventional technology is a method that assumes a use in which inference processing is performed all at once after setting all the prerequisite values, and even if only a part of the values are reset, all The idea is to start the reasoning over again from the beginning. Therefore, if this method is applied to an interactive system that dynamically acquires information during inference, the entire inference will have to be redone each time even if input errors in the previous interaction are corrected, resulting in poor responsiveness. The problem was that it got worse.

The present invention has been made in view of the above circumstances, and its purpose is to solve the above-mentioned problems in conventional reasoning methods, and to make it possible to correct errors in dialog input information in an interactive knowledge base system. The purpose of this invention is to realize and provide an inference method based on a standard mechanism that does not depend on subject knowledge and a method that is highly responsive to dialogue.

[Means for solving problems]

The above-mentioned object of the present invention is to provide a knowledge storage section, an inference processing section that performs inference based on the knowledge stored in the knowledge storage section, an inference situation storage section that stores the inference situation by the inference processing section, and the inference processing section. a means for interacting with a user in order to obtain information necessary for inference by a part, a means for storing a history of updates to the inference situation stored in the inference situation storage part, and a memory for the update history of the inference situation. Means is provided for reversing the inference by referring to the update history stored in the inference situation storage unit and restoring the inference situation stored in the inference situation storage unit to the state before the update, and for each inference cycle, the history of update operations on the inference situation is updated. and the history of interactions with the user, and when instructed by the user, traces this history in order from the newest to the previous, and restores the inference status to the state before the update. This is achieved by an inference method that is characterized by backtracking the inference up to the point of the dialogue.

[Effect]

In the inference method according to the present invention, a knowledge base system having a knowledge storage section, an inference processing section, an inference situation storage section, and an interaction processing section is further provided with an inference situation update history storage section and an inference backtracking processing section.

The inference situation update history storage section is composed of an update information storage section and a dialogue information storage section. Each update information storage section stores information representing the contents of the update operation performed during the corresponding inference cycle, and each interaction information storage section stores information regarding the interaction performed during the corresponding inference cycle. Each,
It is stored in correspondence with the cycle of inference.

The inference backtracking processing unit is activated by the interaction processing unit in response to an instruction from the user, and traces the history stored in the inference status update history storage unit in order from the newest one, according to the contents stored in the update information storage unit. , restoring the inference situation stored in the inference situation storage unit to the state before updating, and determining a position to backtrack the inference according to the content stored in the dialogue information storage unit. The following methods can be used to determine the point of dialogue at which the inference is to be reversed.

(1) The dialog information storage section of the inference status update history storage section stores information indicating the presence or absence of a dialog input during the corresponding inference cycle, and the inference backtracking means, when a user's backtracking instruction is given, While referring to the contents stored in the dialogue information storage section, the process backtracks to the point in time when a dialogue input was made recently.

(2) The dialogue information storage unit of the inference status update history storage unit stores information that specifies the dialogue that took place during the corresponding inference cycle, and the inference backtracking means receives the dialogue point at which the user should backtrack. Obtain the specified information and go back to the specified point while referring to the contents stored in the dialogue information storage section. Furthermore, when obtaining information from the user specifying the point of dialogue to go back to, the history of the specific information of the dialogue stored in the dialogue information storage section may be displayed as a list, and the user may be allowed to specify the point of dialogue in a selection format from the list. good.

As described above, by referring to the storage contents of the update information storage section and the dialogue information storage section of the inference status update history storage section,
While canceling the influence of dialogue input information on the inference situation,
Reasoning can be backtracked to the point of the requested interaction. Since the inference situation is completely restored to the state before entering the dialogue, it is possible to re-execute the inference after changing the input information.

(Example) Hereinafter, an example of the present invention will be described in detail based on the drawings.

FIG. 2 is a configuration diagram of a knowledge base system showing an embodiment of the present invention. In the figure, 201 is a central processing unit that executes various programs and inference processing, etc., 202 is a storage device that stores various programs, data, knowledge, etc., and 203 displays output data of various programs and arrangements for users, etc. 204 is an input device for inputting input data of various programs, user's answers to make-up requests, etc.

FIG. 3 is a functional configuration diagram showing the main parts of the above system.

In the figure, 101 is a knowledge storage unit, 102 is an inference processing unit, 103 is an inference situation storage unit, 104 is an interaction processing unit, and 10
5 is an inference situation update history storage unit, 106 is an inference situation update history storage processing unit, 107 is an inference backtracking processing unit, 110 is an inference situation update processing unit, Ill is a standard interaction processing unit, and 112 is an input/output processing unit. There is.

The functions of each of the above parts will be explained below.

The knowledge storage unit 101 is a part that stores knowledge as information in a format corresponding to its expression method. The method of expressing knowledge includes I
There are various things such as f-then rules, predicate logic, frames, objects, etc. The present invention is applicable to any of these knowledge representation methods, and is also applicable to cases where they are used together.

The inference processing section 102 is a section that performs inference processing based on the knowledge stored in the knowledge storage section 101, and is composed of a knowledge search processing section 108 and a knowledge application processing section 109. Inference methods for knowledge include forward or backward inference of If-Then rules, and derivation of predicate logic.

There are various methods such as searching for frames, sending messages for objects, and the like, and the present invention is applicable to all of them.

The knowledge search processing unit 108 searches the knowledge storage unit 101 for knowledge to be applied next. At this time, if necessary,
If-Then with reference to the contents of the inference situation storage unit 103
Performs matching processing in the case of rules, unification processing in the case of predicate logic, inheritance processing in the case of frames and objects, selection processing in the case of multiple applicable knowledge, etc. The knowledge application processing unit 109 is.

Processing corresponding to the written content of the knowledge searched by the knowledge search processing unit 108 described above is executed.

The inference situation storage unit 103 is a part that stores the inference situation by the inference processing unit 102. The inference situation consists of a collection of various values that express the preconditions, intermediate conclusions, final conclusions, etc. of the inference. Note that in this embodiment, the knowledge storage unit 101
Although an example has been shown in which the system and inference status storage unit 103 are separated, the two may not be clearly separated.

In this embodiment, an inference situation update processing section 110 is provided as a common interface for performing update operations on the inference situations stored in the inference situation storage section 103.

The dialogue processing unit 104 is a part that performs dialogue processing to obtain information necessary for the inference processing unit 102 to proceed with inference from the user. A standard dialogue processing unit 111 is provided as a standard interface for dialogue with users. However, this does not prevent the dialogue processing unit 104 from directly performing a dedicated type of dialogue without going through this interface.

The input/output processing unit 112 performs processing to display the details of the arrangement on the display device 203 when the interaction processing unit 1 (14) or the standard interaction processing unit 111 makes an arrangement to the user, and performs processing to display the content of the arrangement on the display device 203, When obtaining the answer, input device 2
This is the part that performs the process of inputting the answer contents from 04 onwards.

The inference situation update history storage unit 105 is a part that stores the history of updates to the inference situation, and is composed of an update information storage unit 113 and a dialog information storage unit 114. Update information storage section 1
13 and the dialogue information storage unit 114 respectively store information representing the contents of update operations performed on the inference situation and information regarding the dialogue for each inference cycle.

The inference situation update history storage processing unit 106 is a part that performs processing to store the above update history in the inference situation update history storage unit 105.

The inference backtracking processing unit 107 stores the inference status update history storage unit 1
This is a part that refers to the update history stored in 05, restores the inference status to the state before the update, and performs a process of reversing the inference.

The inference of the knowledge base system is performed by the knowledge search processing unit 10.
The process proceeds by repeating an inference cycle consisting of knowledge search by 8 and knowledge application by the knowledge application processing unit 109.

The knowledge application processing unit 109 notifies the inference status update history storage processing unit 106 of the timing for each cycle (signal 116). When the processing content of the knowledge application processing unit 109 is to update the inference status, the update is performed via the inference status update processing unit 110. When the processing content requires interaction with the user, the interaction processing unit 1.04 is activated. In other cases, the corresponding processing section is activated (signal 115).

It is assumed that the dialogue processing unit 104 and other external processing units do not leave side effects that affect the flow of inference in anything other than the driving situation storage unit 103.

Upon receiving the update request, the inference situation update processing section 110 updates the inference situation storage section 103 according to the request, and notifies the inference situation update history storage processing section 106 of the updated contents.

When the dialogue processing unit 104 is started, the standard dialogue processing unit 11
1 to communicate with the user in a standard format (signal 11
9) or engage in direct dialogue in a dedicated format (
signal 117). When updating the inference status based on dialogue input information, the inference status update processing unit 110 is used. The contents of the dialogue are stored in the standard dialogue processing unit 111 or directly by the dialogue processing unit 104 in the inference status update history storage processing unit 10.
6 will be notified.

The inference status update history storage processing unit 106 updates a new update information storage unit 113 and dialogue information storage unit 1 for each inference cycle.
14, and stores the contents of the notified update and the contents of the dialogue.

The inference backtracking processing unit 107 is activated by the dialog processing unit 104 or the standard dialog processing unit 111 that receives the user's backtracking instructions (signals 118, 120) (signals 121, 122).
If necessary, after obtaining an instruction to return to the position (signal 123), the inference situation is returned to the requested dialogue time while referring to the update information storage section 113 and the dialogue information storage section 114.

FIG. 4 is a diagram showing a specific example of description of knowledge based on If-Then rule type expression. The description format follows the knowledge base system construction support tool rES/KERNEL (product name of Hitachi, Ltd.).

In FIG. 4, a rule 300 consists of a condition part 301 and a conclusion part 302. The inference processing unit 102 uses the condition unit 301
When the content of is established, the content of the conclusion section 302 is executed.

The condition section 301 further includes two condition clauses 303 and 304. The conditional clause 303 means that the slot "aspect" of the frame "consultation" is "family situation question and answer." The conditional clause 304 has a class slot of “consulter”.
(305), the slot "marriage" is "married" (306), and the slot "number of children" is rNULLJ (307). However, "?Consultant group" is a variable,
rNULLJ represents an undefined value.

The conclusion part 302 contains the slot “?
This means changing the "number of children" to a value that suits the user. rassignJ is a message that changes the slot value of the frame, and is a message that changes the slot value of the frame, and is sent to the inference status update processing unit 110.
rgetJ is the update command for .

This is a function that returns a dialogue input value, and corresponds to a dialogue command to the standard dialogue processing unit 111.

An example of dialogue by the dialogue processing unit 104 will be explained with reference to FIGS. 5 to 7. FIGS. 5 to 7 are diagrams showing examples of dialog screens.

FIG. 5 shows the screen at a certain point during the question and answer process. 401 is a command area, and 402 is an interaction area. At this point, it is assumed that the user wants to correct the content of the previous answer. There are two ways to return to the corrected part: one is to sequentially return to the previous question, and the other is to directly specify the question part. The corresponding command menu (40
3, 404) with the mouse or function key to start backtracking.

FIG. 6 shows the screen at the time when the direct specification return method is selected and the past question and answer history is displayed.

The system stores the contents of these past question and answers in the dialogue information storage section 114 of the inference status update history storage section 105. Instead of the content of the question and answer itself, the number assigned to the question, the number of one screen and the image of two screens, etc. may be stored and used if the dialog is screen-based.

FIG. 7 shows the screen at the time when the first question is selected on the screen of FIG. 6 and the dialog backtracks to that question.

An example of a table structure for realizing the inference status update history storage unit 105 will be explained with reference to FIGS. 8 to 12.

The inference status update history storage unit 105 is the update information storage unit 11
3 and a dialog information storage section 114, each of which is realized by an update information storage table and a dialog information storage table.

FIG. 8 is a diagram showing the structure of the update information storage table. The update information storage table has an inference cycle ID column 701, an update type column 702, and a frame name column 170.
3, slot name field 704, and slot value field 705
It consists of The inference cycle ID column 701 stores information for identifying the corresponding inference cycle, such as the number of inference stages. The update type column 702 stores information representing the type of update operation.

The frame name 9703 stores information related to the humum name among the accompanying information of the update operation. [4] of the slot name
704 stores information related to the slot name among the accompanying information of the update operation. The slot 1 value 41J705 stores information related to the slot value among the accompanying information of the update operation.

The information storage format of the update information storage table is shown in Figures 9 to 11.
This will be explained using figures. Three types of update operations for the inference situation are assumed: frame generation, frame slot value change, and frame deletion.

FIG. 9 is a diagram showing the storage format of frame generation information in the update information storage table. One frame generation operation corresponds to one piece of table storage information. Table storage information 801 indicates that a frame named "Daichi" was generated in inference cycle "3". In order to restore the state before frame generation, it is sufficient to delete the frame "Daichi", and the necessary information (frame name) for this purpose remains.

FIG. 10 is a diagram showing the storage format of slot change information in the update information storage table. One slot change operation corresponds to one piece of table storage information. The table storage information 901 contains frame “Daichi” in inference cycle “4”.
This indicates that the slot "age" of "rNULLJ" has been changed from rNULLJ to another value. In order to restore the state before changing the slot, all you have to do is change the slot ``age'' of the frame ``Daichi'' to rNULLJ, and the necessary information (
(frame name, slot name, slot value before change) remain.

FIG. 11 is a diagram showing the storage format of frame deletion information in the update information storage table. One frame deletion operation corresponds to several table storage information. The table storage information 1004 indicates that the inference cycle "99" has a class "
This indicates that the frame ``Daichi'' belonging to ``Consultant'' has been deleted. Table storage information 1003 to 1001 shows that before deletion, the slot "number of children" of the frame "Daichi" is "1", the slot "marriage" is "married", and the slot "age" is "30". It represents what happened. To restore the state before the frame was deleted, create a frame ``Daichi'' that belongs to the class ``Consultant,'' and select its slots ``Number of Children'' and ``Marriage.''

It is sufficient to change the "age" to the above value, and the necessary information for that purpose (class name, frame name, slot name 1, slot value before deletion) remains.

FIG. 12 is a diagram showing the structure of the dialogue information storage table. In the dialogue information storage table, inference cycle ID column 1
101, and a dialog specific information column 1102. The inference cycle ID 1illo1 stores information that identifies the corresponding inference cycle. Dialogue specific information column 110
2 stores information specifying the interaction that took place during that inference cycle. If a method of sequentially backtracking to the previous interaction point is used, only information regarding the presence or absence of interaction during the inference cycle is sufficient. When using the method of directly specifying the point of dialogue and returning, information such as the number assigned to the question, the content of the question and answer itself, and the image of the screen number 2 in the case of screen-based dialogue is used.

The processing procedure of the present invention will be explained below with reference to FIGS. 1(,) to (c).

FIG. 1(a) is a flowchart showing the overall processing procedure of this embodiment. First, in step 1201, the contents of the inference situation storage section 103 are referred to as necessary, and knowledge to be applied next is searched from the knowledge storage section 101. When there is more than one applicable knowledge.

Make the selection by any appropriate method. Next, step 120
In step 2, processing corresponding to the searched knowledge description content is executed. If the processing content described in the knowledge instructs the end of the process, it is determined in step 1203 and the process ends. Otherwise, the process returns to step 1201.

FIG. 1(b) is a flowchart showing the procedure of knowledge application processing in this embodiment, and step 1 of FIG.
This corresponds to a detailed version of 202. First, step 13
At step 01, a storage area for the update history corresponding to the current inference cycle is secured in the inference status update history storage unit 105. Next, in step 1302, the processing content described in the knowledge is determined. If it is an interaction process with the user, step 13
Proceed to step 03, and if it is an inference status update process, proceed to step 1.
Proceed to step 304, and if it is other processing, proceed to step 130
Proceed to step 5.

If the process advances to step 1303, execute interactive processing,
In step 1306, the obtained dialogue information is stored in the dialogue information storage unit 114, and in step 1308, it is determined whether there is an instruction to go back during the dialogue. If you have any instructions,
The process advances to step 1310, and if it is not 4, the process advances to step 1309. If the process proceeds to step 1309, update processing is executed based on the information obtained through the dialogue, and in step 1311, the corresponding update information is stored in the update information storage unit 113, and in step 1313, the dialogue processing is completed. Determine whether or not. If the process is completed, the process advances to step 1314; otherwise, the process returns to step 1303.

If the process advances to step 1304, execute the update process,
In step 1307, the corresponding update information is stored in the update information storage unit 113, and the process proceeds to step 1314.

If the process proceeds to step 1305, the corresponding process is executed, and the process proceeds to step 1314.

In step 1314, it is determined whether all processing contents described in the knowledge being applied have been completed.

If it is not finished, the process returns to step l302. If it has ended, the process proceeds to step 1315, where it is determined whether there is any information stored in the update information storage unit 1'13 or the interaction information storage unit 114 during this inference cycle. If so, immediately terminate the process. If there is not, the update history storage area reserved for the current inference cycle is released in step 1316, and the process ends.

If the process proceeds to step 1310, the inference situation updated during the application of the current knowledge is restored, and in step 1312, backtracking processing to a necessary point in time is executed, and the process ends.

FIG. 1(c) is a flowchart showing the procedure of the inference backtracking process in this embodiment, and corresponds to a detailed version of step 1312 in FIG. 1(b). First, in step 1401, the specified backtracking method is determined. If the method is to sequentially go back to the previous dialogue point, the process proceeds to step 1403, and if the method is to directly specify the dialogue point and return, the process proceeds to step 1402.

If the process proceeds to step 1403, the update history is traced back one cycle, and in step 1405 it is determined whether the restoration process for the current cycle has been completed. If it is finished, proceed to step 1409; otherwise, proceed to step 1407
Then, the current cycle is restored based on the contents of the update information storage unit 113, and the process returns to step 1405. Step 14
If the process advances to step 09, it is determined based on the contents of the dialogue information storage section 114 whether or not there was a dialogue in the current cycle. If there has been a conversation, in step 1411, the update history storage area secured from the current cycle to the latest cycle is released, and the process ends. If there is no dialogue, the process returns to step 1403.

If the process proceeds to step 1402, first, an instruction for the return position is obtained, and then, in step 1404, the update history is traced back by one cycle, and in step 1406, it is determined whether the restoration process for the current cycle has been completed. If finished, proceed to step 1410; otherwise, proceed to step 1408
Then, the current cycle is restored based on the contents of the update information storage unit 113, and the process returns to step 1406. Step 14
If the process proceeds to step 10, it is determined based on the contents of the dialogue information storage section 114 whether the current cycle is at the designated return position. If it is the backward position, the update history storage area secured from the current cycle to the latest cycle is released in step 1412, and the process ends. If it is not the backward position,
Return to step 1404.

According to this embodiment, the driving wheel status update FrKI history storage means includes a collection of update information storage units and dialogue information storage units, and each update information storage unit stores update operations performed during the corresponding inference cycle. each dialogue information storage unit stores information representing the content, and each dialogue information storage unit stores information regarding dialogues conducted during the corresponding inference cycle in correspondence with the inference cycle;
The inference backtracking means is activated by the interaction means in response to an instruction from the user, and traces the history stored in the inference status update history storage means in order from the newest one, according to the contents stored in the update information storage unit: Restoring the inference situation stored in the situation storage unit to the state before updating, and determining the position to backtrack the inference according to the memory contents of the dialogue information storage unit, thereby reducing the influence of dialogue input information on the inference situation. By configuring the system to backtrack the inference to the requested dialog point while canceling it, and re-execute the inference after changing the input information, errors in dialog input can be corrected in an interactive knowledge-based system. This has the effect that an inference method can be realized using a standard mechanism that does not depend on subject knowledge and a method that is highly responsive to dialogue.

The dialogue information storage section of the inference status update history storage means stores information indicating the presence or absence of dialogue input during the corresponding inference cycle, and the inference backtracking means refers to the stored content of the dialogue information storage section. An inference method configured to allow inference to be sequentially backtracked to the previous dialog point each time there is a backtracking instruction by backtracking to a point in time when there was a recent dialog input, or the inference status update history storage. The dialogue information storage section of the means stores information specifying the dialogue that took place during the corresponding inference cycle, and the inference backtracking means obtains information specifying a dialogue point to go back to from the user, and stores the dialogue information. It is also possible to carry out an inference method in which the user can directly specify the point in time in which the user wants to go back in reasoning by backtracking to a specified point in time while referring to the memory contents of the storage unit. Furthermore, the inference method is such that the inference backtracking means displays a list of the history of specific information of the dialog stored in the dialog information storage section of the inference status update history storage unit, and allows the dialog point at which the inference is to be backtracked to be specified in a selective format. is also possible.

It goes without saying that the above embodiment is shown as a preferred embodiment of the present invention, and that the present invention is not limited thereto.

〔Effect of the invention〕

As described above, according to the present invention, a knowledge storage section, an inference processing section that performs inference based on the knowledge stored in the knowledge storage section, an inference situation storage section that stores the inference situation by the inference processing section, A knowledge base system comprising means for interacting with a user to obtain information necessary for inference by the inference processing unit, a means for storing a history of updates to the inference situation stored in the inference situation storage unit, and the inference situation. A means is provided for reversing the inference by referring to the update history stored in the update history storage means and restoring the inference situation stored in the inference situation storage unit to the state before the update, and updates the inference situation for each inference cycle. It memorizes the history of operations and the history of interactions with the user, and when instructed by the user, restores the inference status to its pre-update state by following this history in descending order. Since the inference is made to backtrack until the point of the requested interaction, an inference method that can correct errors in the interaction input in an interactive knowledge-based system can be combined with a standard mechanism that does not depend on the target knowledge. This has the remarkable effect of being realized by a highly responsive method.

[Brief explanation of the drawing]

Figures 1 (a) to (c) are flowcharts showing processing procedures according to an embodiment of the present invention, Figure 2 is a configuration diagram of a knowledge base system according to the embodiment, and Figure 3 is a functional configuration showing the main parts thereof. Figure 4 shows a specific example of knowledge description based on If-Then rule type expressions, Figures 5 to 7 show examples of dialog screens, and Figure 8 shows the update information storage table. 9 to 11 are diagrams showing the information storage format of the update information storage table, and FIG. 12 is a diagram showing the structure of the dialogue information storage table. 101: Knowledge storage section, 102: Inference processing section, 1o3: Inference situation storage section, 104: Dialogue processing section, 105: Inference situation update history storage section, 106: Inference situation update history storage processing section, 107: Inference backtracking processing section , 108: Knowledge search processing unit, 109: Knowledge application processing unit, 110: Inference status update processing unit, 111: #J quasi-dialogue processing unit, 112: Input/output processing unit, ] 13: Update information storage unit, 114: Dialogue information storage unit,
115: Applicable process start, 116: Inference cycle notification, 1
17: Dedicated dialogue, 118, 120: Backward instruction,
119: Standard dialogue, 121, 122: Start of inference backtracking process, 123: Backtracking position instruction, 201: Central processing unit, 202: Storage device, 203: Display device. 204: Input device, 401: Command area of dialogue screen,
402: Same dialog area, 403, 404: Same command menu, 701: Inference cycle ID column of update information storage table, 702: Same update type column, 703: Same frame name column, 704: Same slot name column, 705: Same Slot 1 to value column, 1101: Inference cycle ID column of dialog information storage table, 1102: Same dialog specific information column. Figure 1 (a) Figure 1 (c) Figure 2 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11

Claims (1)

[Scope of Claims] 1. A knowledge storage unit, an inference processing unit that performs inference based on the knowledge stored in the knowledge storage unit, an inference situation storage unit that stores inference situations by the inference processing unit, and the inference processing unit In a knowledge base system having means for interacting with a user to obtain information necessary for inference by a processing unit, means for storing a history of updates to the inference situation stored in the inference situation storage unit, and an update history of the inference situation providing means for reversing the inference by referring to the update history stored in the storage means and restoring the inference situation stored in the inference situation storage unit to the state before the update;
For each inference cycle, the history of update operations for the inference status and the history of interactions with the user are memorized, and when instructed by the user, these histories are followed in order from the newest to the newest. An inference method characterized by restoring the inference situation to the state before updating and backtracking the inference to the requested dialogue point. 2. The inference status update history storage means includes a collection of update information storage units and dialogue information storage units, and each update information storage unit stores information representing the contents of update operations performed during the corresponding inference cycle, and , each dialogue information storage unit stores information regarding dialogues performed during the corresponding inference cycle in correspondence with the inference cycle, and the inference backtracking unit stores information regarding the dialogue performed during the corresponding inference cycle, and the inference backtracking unit stores information regarding the dialogue performed during the corresponding inference cycle, and the inference backtracking unit stores information regarding the dialogue performed during the corresponding inference cycle, and the inference backtracking unit stores information regarding the interaction performed during the corresponding inference cycle, and the inference backtracking unit and restores the inference situation stored in the inference situation storage unit to the state before the update according to the storage contents of the update information storage unit while tracing the history stored in the inference situation update history storage unit in order from the newest one. and determining a position to backtrack the inference according to the memory contents of the dialog information storage unit, thereby backtracking the inference to the requested dialog point while canceling the influence of the dialog input information on the inference situation; 2. The inference method according to claim 1, wherein the inference method is configured to re-execute the inference after changing input information. 3. The dialogue information storage section of the inference status update history storage means stores information indicating the presence or absence of dialogue input during the corresponding inference cycle, and the inference backtracking means refers to the storage content of the dialogue information storage section. Then, go back to the point where there was the most recent dialogue input, and every time there is a backward instruction,
The inference method according to claim 2, characterized in that the inference method is configured so that the inference can be sequentially backtracked to the point of the previous interaction. 4. The dialogue information storage section of the inference status update history storage means stores information specifying the dialogue that took place during the corresponding inference cycle, and the inference backtracking means specifies a point in time of the dialogue to go back to from the user. The present invention is configured so that the user can directly specify the dialogue point at which he or she wishes to go back to the inference by obtaining information about the interaction information and backtracking to the designated point in time while referring to the memory contents of the dialogue information storage unit. characterized by
An inference method according to claim 2. 5. The inference reversing means displays a list of the history of the specific information of the dialogue stored in the dialogue information storage section of the inference status update history storage means, and makes it possible to select and specify the dialogue point at which the inference is to be reverted. The inference method according to claim 4, characterized by: