JPH096620A - Rule generation support tool for expert system - Google Patents

Abstract

PURPOSE: To enable a rule generator to easily recognize the independence relations between rules at the time of generating rules with a text type tool. CONSTITUTION: When rules of an expert system are generated by a text type tool generation support tool, a rule table where rule names of all rules and their condition parts and execution parts are divisionally generated and preserved is prepared. When a rule name, a retrieval direction, and a pattern are inputted (S1), it is checked in the retrieval direction whether the rule has an effect or is affected (S2), and execution parts in the table are retrieved by the pattern in the case of the backward direction (S3), and the collated rule is placed on the left of the noticed rule to display a tree structure on a picture (S4). Condition parts in the table are retrieved by the pattern in the case of the forward direction (S5), and the collated rule is placed on the right of the noticed rule to display the tree structure on the picture (S6). Thus, the rule generator can intuitively understand the dependence relations between rules.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an expert system rule making support tool, and more particularly to a text type rule making support tool.

[0002]

2. Description of the Related Art The expert system basically consists of
It consists of a knowledge base that stores the knowledge of experts as rules and an inference engine that solves problems by referring to this knowledge base.

When constructing an expert system, a rule (expressed in the form of "if condition part then execution part") in which expert knowledge is converted into an expression that can be processed by a computer is roughly divided into a graphic. There are two types of support tools, type and text type.

The graphic type is a support tool capable of creating rules such that a picture is drawn in consideration of connections between rules when creating rules. That is, as shown in FIG. 7, when a rule execution part affects a condition part of another rule by using a dedicated graphic tool, a connection between them is drawn in a tree structure using a mouse.
Create a rule by entering the condition part and execution part for the rule specified with the mouse.

In the text type, the rules are created by fragmentally expressing the text in the if-then-form without paying attention to the connection between the rules. That is, as shown in FIG.
Using a text editor, enter the condition part and execution part in text for each rule.

In this text type, the rule expressed in text is often managed by a text editor, and even in the rule maintenance function created exclusively for the expert system, the rule expression is written in Japanese as shown in FIG. There is nothing more than that.

[0007]

The advantage of the rules of the expert system is that its fragmentary description is sufficient. In other words, being aware of the relationship with other rules conflicts with the goodness of the system.

In the conventional graphic type tool, since the rules are created while being aware of the connection between the rules, an action contradicts the goodness of the system, and the goodness of the rules cannot be utilized. Furthermore, in the graphic type, when rules are complicatedly dependent on each other, as shown in FIG.
There is a problem that the visual effect of understanding rule relationships is reduced.

With respect to these points, the text type tool takes advantage of the goodness of the rule, but the management of the text type rule is effective in the graphic type tool "easy understanding of rule relations". On the contrary, there is a problem that "rules are not easy to understand".

For example, when modifying or adding a rule, if one wants to know another rule that influences the rule or another rule that the rule affects, as shown in FIG. 11, use a text editor. You have to find out all the applicable rules by performing a character string search of a text file in which the rules are written.

At this time, if the search or correction / change of the corresponding rule fails, the dependency between the rules may be broken.

It is an object of the present invention to provide a rule creation support tool which can easily know the dependency relation between rules in rule creation by a text type tool.

[0013]

In order to solve the above-mentioned problems, the present invention has a function of text-creating a rule to be stored in a knowledge base of an expert system in the form of "if condition part then execution part". In the provided text-type rule creation support tool, enter a rule table that creates and saves all created rules by dividing them into rule names, their condition parts and execution parts, and the specified rule name, search direction, and pattern And a matching rule display function for searching a rule affected by the rule name or a rule affected by the rule name from the rule table according to the search direction and pattern, and displaying the rule name tree structure on the screen. It is characterized by having.

[0014]

[Function] When creating rules in the text type, the rules that have a dependency relationship with the specified rule are searched from the rule table and displayed in a tree structure on the screen. It is expressed as a tree diagram so that rule authors can intuitively understand the rule relationships.

[0015]

[Embodiment] A text-type rule creation support tool uses a text editor, which is a conventional basic function, to enter text for the condition part and execution part for each rule, and to manage the rules expressed in text. In addition to the function performed by the editor, a collation rule display function for visually displaying the dependency between rules is provided.

FIG. 1 is a flow chart for realizing the collation rule display function. The rule table used in this matching rule display function is created and saved in advance for all the rules created so far by dividing the rule name, its condition part, and execution part. As shown in, it consists of a rule name, a condition part and a text string of an execution part.

In FIG. 1, in order to display a rule having a dependency relationship, first, in step S1, a rule name of interest, a search direction and a pattern are determined and input.

Of these, the search direction determines whether the rule of interest is to be examined for other rules that are affected (backward rules) or for other rules that are affected (forward rules). . The pattern determines the patterns that affect each other among the rules.

FIG. 3 shows an example of an input screen in the process S1, in which an input area for a rule name, a search direction and a pattern is prepared in the search rule dialog. For the rule name, enter the name of the rule of interest, and specify whether the direction is the direction in which the rule affects (forward) or the direction in which it affects (backward). For the pattern, if the search direction is forward, enter the pattern that is considered to be in the condition part of another rule, and if the search direction is backward, enter the pattern that is considered to be in the execution part of another rule.

Up to four patterns can be input, for example, under a logical product condition. Also, "*" is used as a wildcard, which means it matches any word.

In the input example of FIG. 3, paying attention to rule 1, it indicates that the search is performed in the forward direction. The pattern is that rule 1 of FIG. 2 turns on the slot “ramp” of “frame” in the execution unit. Therefore, a pattern is searched for a rule having a conditional part that a lamp of a certain frame (FRAME) is on.

After the determination in the process S1, the collation rule display function checks whether the search direction is the forward rule or the backward rule (S2), and when the backward rule is searched, the execution unit of the rule table is searched. Search for the pattern (S
3) If there is a rule matched by this search, the rule name of interest and the affected rule to the left of it are displayed in a tree form on the screen (S4).

FIG. 4 exemplifies a tree display screen when a backward rule is searched, and rules 1 and 10 are displayed as rules affected by rule 5.

When the forward rule is searched in check S2, the pattern of the condition part of the rule table is searched (S
5) When there is a rule matched by this search, the rule name of interest and the rule affecting it to the right are displayed in a tree form on the screen (S6).

FIG. 5 exemplifies a tree display screen when a forward rule is searched, and rules 3, 5, and 6 are displayed as the rules affected by rule 1.

In the above-described processes S2 to S7, only the rules adjacent to the focused rule are checked, and in order to check the destination of the adjacent rule, the adjacent rules are used as new focused rules to perform the processes S2 to S6 (S7). ).

FIG. 6 shows an example of a screen display when two searches are performed. When the backward search of the rule 5 is performed and then the adjacent rules 20 and 22 are displayed by the backward search of the adjacent rule 10. Is.

Therefore, according to the present embodiment, it is possible to quickly and accurately grasp the dependency relationship between rules, which was difficult with the table type tool, by executing the collation rule display function.

That is, in order to know the dependency relation between the rules of the text expression, if a person tries to find out from the existing rule group, the character strings of all the rules must be read for a long time, and the oversight can be overlooked. It often happens.

On the other hand, if the collation rule display function is used, the search becomes quick and accurate. Also, since the dependency relationship is represented by a tree-like diagram, it becomes possible to intuitively understand the dependency relationship.

[0031]

As described above, according to the present invention, when a rule is created in a text type, a rule having a dependency relationship with the specified rule is searched from the rule table and displayed on the screen in a tree structure. Since it is provided, it is possible to quickly and accurately grasp the dependency relationship between rules, which was difficult with the rule of text expression, and it is easy to intuitively understand.

[Brief description of drawings]

FIG. 1 is a flowchart of a matching rule display function according to an embodiment of the present invention.

FIG. 2 is a rule table according to an embodiment.

FIG. 3 is a search rule dialog according to the embodiment.

FIG. 4 is a rule tree display screen at the time of backward search in the embodiment.

FIG. 5 is a rule tree display screen at the time of forward search in the embodiment.

FIG. 6 is a rule tree display screen when two searches are performed in the embodiment.

FIG. 7 is an example of a graphic type rule creation screen.

FIG. 8 is an example of a text type rule creation screen.

FIG. 9 is an example of a text-type Japanese rule creation screen.

FIG. 10 is an example of a display screen of rules that are complicatedly dependent on each other.

FIG. 11 is an example of a character string search screen by an editor.

Claims (1)

[Claims] 1. In a text-type rule creation support tool having a function of creating a text representation of rules stored in a knowledge base of an expert system in a format of “if condition part then execution part”, all created rules A rule table that creates and saves a rule name and its condition part and execution part separately, and when the specified rule name, search direction, and pattern are entered, the rules that the rule with that rule name affects A rule creation support tool for an expert system, comprising: a matching rule display function for searching an affected rule from the rule table according to the search direction and pattern and displaying the rule name on a screen in a tree structure.