JPH08147168A - Knowledge verification supporting method for knowledge processing system - Google Patents

Abstract

PURPOSE: To easily perform a debugging operation by speedily investigating the cause why a rule to be ignitted is not ignited differently from the expectation of a user when trouble not to ignit that rule is generated as to a non-ignited rule to be an object. CONSTITUTION: In the case that an object expected as an igniting object frame is not stored in a memory node, through a company F is expected to satisfy the conditions of a condition node 1 of a rule 1 (the company F is expected to be described in a memory node 613), the company F is not stored in the node 613. Therefore, the user selects the node 613 while using a mouse, selects the company F from a frame list display area 606, instructs a knowledge verify button 605 and starts knowledge verification. A system recognizes the failure of collation by comparing a node 612. From this result, the node 612 is emphatically displayed, and a condition term corresponding to this node is emphasized by the rule just displayed in a rule display area 601.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a knowledge verification support method for a knowledge processing system, and in particular, in the knowledge debugging work in the process of constructing a knowledge processing system, it is different from a user's expectation when debugging unfired rules. The present invention relates to a knowledge verification support method suitable for investigating the cause.

[0002]

2. Description of the Related Art Knowledge processing systems aim to "solve problems at the same level as experts by porting their specialized knowledge and know-how to a computer." 198, aiming at expansion, efficient and stable operation, and improvement of control quality
It has been actively applied since the 0's.

"Knowledge" in a knowledge processing system is generally referred to as "frame" which represents the facts of the target world, as described in "Knowledge Representation and Utilization" (published by Ohmsha Haruki Ueno and Mitsuru Ishizuka). If-t that represents the rules of the target world (what judgments and directives to make when facts are in a particular state)
It consists of "rules" in hen form. The process of deriving the final fact from the rule and the fact of the initial state is called "inference", and the inference by the rule and the frame is performed as follows. First, it is checked which condition part of the rule is satisfied in the initial frame state (this process is called "matching"). 1 from the rules that satisfy the condition part (when the condition is satisfied is called "firing" and that rule is called "firing rule")
Select one rule (this process is called "conflict resolution"). Execute the conclusion part of the rule (this process is called "execution"). In the rule conclusion part, the contents of the frame are changed and the frames are added / deleted, so that the collation is performed again. In this way, inference is performed through a cycle of collation, conflict resolution, and execution.

As can be seen from this inference processing, the order of processing in the knowledge processing system is the execution order of the rules itself, and is dynamically determined according to the data change of the frame and is not described in advance. Further, depending on how the rule is described, it is possible to search for a frame satisfying the condition, not only checking whether the frame satisfies the rule condition part, and dynamically determine the frame to be processed.

On the other hand, in the processing by the conventional procedural language, "which data, in what order and how to process" is programmed in advance, and the processing is only executed in accordance with this. . There is no need for “verification” or “conflict resolution”.

As described above, the knowledge processing system and the conventional procedural language processing have completely different processing steps, and the debug (knowledge verification) method also necessarily differs.

Generally, in debugging, in the case where the output value of the processing result is erroneous with respect to the input value, the processing process is traced, and the logical value or the data defect is found while confirming the data value. . In the case of procedural language processing, "what data is processed, in what order, and how" is programmed in detail, and traces can be made for each detailed processing unit. Therefore, if the contents of the program are known, it is possible to relatively easily find the defective portion only by tracing and checking the data.

On the other hand, in the knowledge processing system, it is not easy to find the defective portion only by tracing the rule execution and checking the frame contents. The unit of the inference trace in the knowledge processing system is a rule like the unit of execution, and it is possible to confirm in what order the rules are executed by the trace. Therefore, the inference trace may cause a rule that the user intended to "execute" not to be fired without being fired, or that a rule intended to "fire only once" would be executed many times. Can be confirmed.

However, in order to pursue the cause, which of the plurality of conditional clauses described in the rule conditional section does not fire without being satisfied, or which frame combination causes the rule to fire. I have to find out what I am doing. In other words, collation and confirmation of conflict resolution are important tasks for finding defective parts in the knowledge processing system.

A conventional knowledge processing system construction support tool described in, for example, Japanese Patent Laid-Open No. 4-32931 has a function of displaying a competitive set (a firing rule and a frame combination), and a rule "it should not fire" is actually used. Can check which frame combination is firing.

[0011]

In the conventional technique, only the rule that fired is targeted for debugging. However, there is no function to check the matching status for rules that have not fired, and if the rule that the user intends to "fire" does not fire, visually compare the content of the rule and the content of the frame, Since the condition clause or which frame value caused the fire not to be found was discovered, there was a problem that debugging was very troublesome.

That is, as described above, in the conventional method, when a problem that "a rule that should fire is not fired" occurs, the contents of the rule and the contents of the frame that should be fired are visually checked to prevent firing. You have to find out the cause. In particular, in a rule with a large number of conditions or in a rule for searching for a frame satisfying a condition, there is a problem that visual confirmation by a user is frequent, it takes time to find a defective portion, or a defective portion is overlooked.

The present invention has been made in order to solve the above-mentioned problems of the prior art, and its object is to debug unknowing rules in a knowledge processing system for describing knowledge by rules and frames. In the event of a problem that a rule that should fire does not fire, a cause that does not fire that rule, which is different from the user's expectations, can be quickly investigated and a knowledge verification support method that facilitates debugging work can be provided. It is in.

[0014]

In order to achieve the above object, the structure of the invention relating to the knowledge verification support method of the knowledge processing system according to the present invention is such that the “fact” regarding the target problem area is framed and A knowledge processing system for performing inference by describing knowledge based on a "judgment rule" that guides another fact, which comprises inference means, means for performing knowledge verification, means for storing the frame and the rule, and knowledge. In a knowledge verification support method for a knowledge processing system, which comprises a state display means, the frame to be subjected to knowledge verification, and the input means for specifying the rule and giving a command to the system, a frame being inferred to the knowledge state display means. The state and the collation state of the rule are displayed, and the frame and rule specified by the user by the input means are targeted for knowledge verification. Finding the cause that the specified rule has not fired using the knowledge verification means, and in the knowledge state display means, the condition part and frame of the rule that caused the rule to fire The value of and is explicitly displayed.

In order to achieve the above object, another structure of the invention relating to the knowledge verification support method of the knowledge processing system of the present invention is that the rule has at least one frame in which a part of the conditions is not satisfied. When the rule itself is described as not valid, and when this rule is subject to knowledge verification, the knowledge verification means is used to identify a frame in which a part of the conditions is not satisfied. This frame is clearly displayed on the knowledge status display means by finding out the cause of the ignition.

More specifically, in the knowledge verification support method for the above knowledge processing system, a rule is expanded and a comparison node for performing a condition comparison process corresponding to each condition of the judgment rule and a condition of the judgment rule are partially included. In the case where the inference is performed by using the collation processing network including the memory node that performs the storage processing of the facts satisfied with the above, the structure of the collation processing network is displayed on the knowledge state display means as a network structure diagram, and the network structure is displayed. The collation state of the inference result is displayed on the figure.

More specifically, the frame stored in the memory node is displayed on the network structure diagram, the frame is specified by the input means, knowledge verification is performed as a target of knowledge verification, and collation fails. The comparison node corresponding to the condition part of the rule is specified on the network structure diagram.

[0018]

According to the present invention, the value of the memory node is displayed on the screen by displaying the network used for the inference processing in which the rule is expanded. Therefore, the user starts debugging by designating the frame stored in the memory node when the result is not intended by the user. Then, the condition part of the rule for which the matching has failed and the slot of the frame are clearly displayed. Therefore, by simply specifying the rule that the user expects to fire and the frame that should trigger the fire, the defect can be found instantly, improving debugging efficiency compared to the past and improving the knowledge processing system. Development man-hours are also shortened.

Further, in the rule including the absolute condition, the frame which does not satisfy the condition can be easily found at the time of debugging, which also improves the debugging efficiency.

Since the matching process is limited to the specified rule and the specified frame and the matching process is locally performed, the side effect of changing the matching state of the whole inference does not occur.

[0021]

Embodiments of the present invention will be described below with reference to FIGS. 1 to 17. [Embodiment 1] Hereinafter, each embodiment according to the present invention will be described with reference to FIGS. 1 to 12. (I) Knowledge Verification Support System Configuration First, the system configuration of the knowledge verification support method of the present invention will be described with reference to FIG. FIG. 1 is a block diagram showing the configuration of a knowledge processing system having knowledge verification means.

The general structure of this system is divided into an inference section, a knowledge verification support section, and an input means. The inference unit 110 includes an inference unit 100, a frame storage unit 101, a rule net storage unit 102, and a rule storage unit 103. The "rule net" here means
In order to perform inference at high speed, rules are developed in the form of a discrimination network, and the Rete network and the like are famous (for reference, for example, "Rete: A Fast A
lgorithm for the Many Pattern / Many Object Patter
n Match Problem ", Artificial Intelligence vol.19, N
o.1, pp.17-37 (1982)). Then, the inference processing that proceeds in the cycle of collation-conflict resolution-execution is performed by the inference means 100 using the frame in the frame storage means and the rule net in the rule net storage means.

The knowledge verification support unit 111 comprises a frame display means 104, a rule net display means 105, a rule display means 106, an inference control means 107, a rule net information storage means 108, and a knowledge verification means 109.

The rule net information storage means 108 stores information for actually displaying the rule net in a network form on the screen and information necessary for local collation for finding a defective portion. Here, the “defective portion” is a portion where a user makes an inference unintended for debugging the knowledge processing system.

The inference control means 107 controls the inference means 100 and realizes step execution of inference, which is a work necessary as the first step of finding a defective portion, and execution by breakpoints.

The knowledge verifying means 109 actually performs local matching while controlling the inference means 100 based on the local matching information in the rule net information storing means 108.
The defective portion is specified, and the defective portion is clearly shown to the user through the rule net display means 105.

For this knowledge verification, a user instruction or the like is given from the input means 112.

(II) Knowledge and Rete Network This embodiment is premised on a knowledge processing system that performs inference processing by the above-mentioned Rete network.
Therefore, next, the principle of the Rete network and the rules and frames used in the description of this embodiment and Rete
An example of a network will be described with reference to FIGS. FIG. 2 is a diagram showing rules used in the first embodiment according to the present invention. FIG. 3 is a schematic diagram in which the rules shown in FIG. 2 are expanded and shown in the Rete network. FIG. 4 is a schematic diagram showing an expression on a memory cell of the Rete network.

FIG. 5 is a view showing a frame used in the first embodiment according to the present invention.

In the knowledge processing system, the rule is used as a "rule" for inference as described above, and the description grammar is defined so that the system can interpret the rule.
FIG. 2 is written according to the grammar thus defined.

The symbol used in FIG. Is a symbol used at the beginning of the variable, for example, "? Company 1"
Represents a variable with a variable name of company 1. At the time of matching, any variable is matched, and a value is assigned to the variable. In the same rule, the same variable name means the same variable. @
Is a symbol representing a slot of a frame (a place where a value is stored). For example, “C company @ capital” represents a slot of capital of a frame of C company.

The rule is divided into an if part and a then part. Of these, the if part represents a condition, and a plurality of conditions are combined and described by an AND condition. the
The n part describes the operation to be executed when the condition is satisfied. In the example of this FIG. 2, it is interpreted as follows.

Assume that the "capital" slot value of company C is Y and the "number of stores" slot value is Z. "Listed" slot value is 1
Part 1, the frame of “Profit” slot value is 100, and the frame of “Number of stores” slot value is Z (that is, coincides with “Number of stores” slot value of company C) is company 1. Let X be the "credit rank" slot value of the company 1.

The "listed" slot value is 2 copies, the "credit rank" slot value is X (that is, it matches the "credit rank" slot value of Company 1), and the "capital" slot value is Y (that is, The frame of “Capital” slot value of company C) is set as company 2.

That is, the rule shown in FIG. 2 is that "company 1 is a company that satisfies the condition that the profit is 10 billion in the TSE First Section listed company and the number of stores is equal to C company, and TSE 2
Company 2 is a listed company that satisfies the condition that the capital is equal to C company. Search for a combination with the same credit rank among the combinations of company 1 and company 2, and then the
Do "n or less."

FIG. 3 shows this rule developed in the Rete network. Table 1 below shows the roles of the symbols used in each node, their names, and their functions.

[0037]

[Table 1]

According to the structure of the network shown in FIG. 3, the relationship between the frame and the rule is checked. The intra node and the inter node perform a condition comparison (these are referred to as "comparison node") and an α memory node is used. The β memory node serves as a node for storing a frame satisfying the conditions (both of which are called “memory node”), and the role of the node is divided into two. This aims at high-speed processing in anticipation of a frame state change. That is, since the collation state is stored in the α memory and the β memory, it is not necessary to collate all rules and all frames every collation process.
Only the frames whose slot contents have changed need to be re-matched.

Therefore, the frame in which the contents of this slot have changed is flowed downward from the root node, and the comparison and storage processing is performed in each node by the function shown in Table 1. It should be noted that performing the comparison / storage processing on the frame according to the structure of the Rete network is referred to as "flowing" the frame to the network.

When the comparison is successful in the inter node or the intra node, a frame combination including a changed frame is created, the frame is stored in each α memory node or β memory node, and the frame is further flowed downward.

When a change frame or a combination of change frames reaches a terminal node corresponding to a terminal of the network, the rule corresponding to the network is fired.

In addition, while streaming to the network,
When the condition comparison is not satisfied, the change frame is stopped from flowing downward, and the node is traced back to the branch branch which has not been flowed yet. When the change frame has finished flowing through the network, the matching process is complete.

Note that the Rete network expansion example shown in FIG. 3 relates to only one rule, but in the case of a multi-rule network expansion example, the network of each rule hangs below the root node. It will be. In addition, even if the rules are different, if there is a common condition comparison part, it is put together to improve the speed.

This Rete network is generally held as a structure in which memory cells are chained on a memory as shown in FIG.

Hereinafter, in this embodiment, the rule shown in FIG. 2 and the Rete network shown in FIG. Then, as the frames, the six frames shown in FIG. 5 will be used for description. this is,
It is a model of a company and its attributes.
"Profit", "Credit rating", "Capital", "Number of stores"
Slots such as correspond to those used for evaluation of an actual company.

(III) Procedure of Knowledge Verification Support Method and Its User Interface Next, the procedure of the knowledge verification support method according to the present invention and its user interface will be described with reference to FIGS. 6 to 12. First, an approximate procedure of the knowledge verification support method according to the present invention will be described in the order of FIG. Figure 6
[Fig. 6] is a general flowchart showing a procedure of the knowledge verification support method according to the present invention.

It is assumed that the user sees the result of the inference and recognizes a defect that the rule that should fire does not fire. At this time, the user firstly makes the inference control means 107.
The step execution and break execution of the inference are performed by, and the inference is temporarily stopped when the rule that should fire is not fired (S01).

Then, the user displays the rule display means 106.
From the rules displayed on the screen, a rule which is supposed to fire is designated from the input means 112, and a frame stored in the rule network information storage means 108 which should cause firing of the rule is designated (S0).
2).

Next, in the system, the knowledge verification means 109 performs local collation by using the designated rule and frame, and identifies the defective portion (that is, the portion where the collation fails) (S03).

Finally, the identified defective portion is clearly indicated by the rule network display means 105 (S0
4).

Next, the user interface of this knowledge verification support system will be described with reference to FIG. Figure 7
FIG. 3 is a diagram showing an example of a screen display of a knowledge verification support system according to the knowledge verification support method of the first embodiment of the present invention (No. 1).

As information for displaying the rule net as shown in FIG. 7 in a network form, the system creates the information as shown in FIG. 4 for each rule together when the rule net is created. Information storage means 108
Need to remember. As the rule net information storage at this time, the number of conditional clauses, the number of intra nodes for each conditional clause, the addresses of the intra nodes and the α memory nodes (the mnemonic address of the rule net code stored in the rule net storage means 102) , The number of internodes where the conditional clauses join, the internode and β memory node address, etc.

In the screen shown in FIG. 7, as a display area, a network display area 604 of rule net, a rule display area 601, and frame content display areas 602 and 60 are displayed.
3, a frame list display area 606, a content display instruction button 607 for inputting a command to the system, and a defect finding button 605.

The frames to be debugged are displayed as a list in the frame list display area 606. To make them a target of operation, operate the mouse cursor with the mouse and point to the frame column. Good.
The designated column is highlighted with high brightness like [Company F] in FIG. 7.

When starting the knowledge verification, the user may point the defect finding button 605 with the mouse.

Network display area 60 of rule net
In 4, the Rete network of the rule to be debugged is displayed. What is displayed in this FIG. 7 is the Rete network of FIG. 3, and the symbols of the displayed nodes also correspond to those described in Table 1.

That is, in this rule net expression, a double square means a comparison node (intra node and inter node). By pointing this symbol with the mouse and selecting the content display button 607, the rule display area 601 is displayed.
The condition corresponding to the node picked with the mouse of the rule displayed in is highlighted with high brightness. In FIG. 7, the node 612 corresponding to the node 2 in FIG. 3 is highlighted. What is the corresponding condition? This means that Company 1 has 100 "profit" slots.

In this rule net expression, double circles are memory nodes (α memory node and β memory node).
Means When this symbol is designated with the mouse and the content display button 607 is selected, the content of the memory is pop-up displayed based on the node address stored in the rule net information storage means 108. In FIG. 7, the node 613 corresponding to the node 4 has ["A company"] ["D
Company "], [" C to node 617 corresponding to node 10
Company ”], node 619 corresponding to node 7 [(” A
“Company”, “Company B”)] [(“Company D”, “Company B”)] are respectively displayed.

As described above, the ability to confirm the contents of the memory node is a useful means for narrowing down the reason why the firing target frame or the combination of frames has not been made.

Now, with reference to FIGS. 3 and 6, how to specifically perform knowledge verification will be described with reference to FIGS. 7 to 12. FIG. 8 is a diagram showing a screen display example of the knowledge verification support system according to the knowledge verification support method of the first embodiment of the present invention (No. 2). Figure 9
3 is a flow chart showing an algorithm for finding a defect in case 1 of searching an intra node (part 1). FIG. 10 is a flowchart showing an algorithm for finding a defect in case 1 of searching an intra node (part 2). FIG.
FIG. 1 is a flowchart showing an algorithm for finding a defect in case 2 of searching an internode (part 1). FIG. 12 is a flowchart showing an algorithm for finding a defect in case 2 of searching an internode (part 2).

This knowledge verification system is aimed mainly at finding a defect in the rule that the user does not fire, but there are generally two cases in which the rule does not fire. The first case is (1) what is expected to be the firing target frame is not stored in the α memory node, and the second is (2) what is expected to be the firing target frame is the α memory node. However, it is not stored in the β memory node.

Each case will be described below with reference to examples.

(1) Case in which what is expected as the frame to be fired is not stored in the α memory node The user said, “Company F satisfies the condition of Condition 1 of Rule 1 (Company F is the memory node. 613)) ”. However, as can be seen from the display of FIG. 7, company F is not stored in the node 613.

Therefore, in order to know the cause, the user selects the memory node 613 with the mouse, selects company F from the frame list display area 606, and instructs the knowledge verification button 605 to start the knowledge verification. As a result, the knowledge verification unit 109 is activated, and the system recognizes that the collation has failed in the comparison of the node 612 corresponding to the intranode node 2. (That is, the value of the “profit” slot of Company F is 80.) Therefore, based on this result, the node 612 is highlighted, and the condition term corresponding to this node is displayed in the rule display area 601. To emphasize. Further, the contents of company F are displayed in the frame display area 602, and the “profit” slot, which is the compared slot, is highlighted. These displays make it clear to the user that the cause of the failure was a comparison of the "Profit" slot between Company F and the first condition clause.

In this way, to find a defective part,
The algorithm shown in FIG. 9 can be used. First, the specified α memory node is searched for information in the rule net information storage means 108, and all the node addresses of the intranodes above it are retrieved (S80).
1).

Next, the changed frame to be sent to the rule net is set as the designated frame (S802). Then, the system for each extracted intranode is the inference means 10
0 is started and the conditions are compared (S803, S80
4, S805).

If the condition comparison fails, the intra-node causes a trouble and the intra-node is taken out (S806).

Further, when it is desired to find a plurality of defective points at the same time, the algorithm shown in FIG. 9 can be used.

That is, even if a defective intranode is found (S806), the processing is not ended there, but a loop may be made to perform the next condition comparison (S80a).

In this way, it is possible to know more defective parts by one debugging operation, and it can be expected to contribute to the progress of the debugging work by the user.

(2) Case in which the expected frame to be fired is stored in the α memory node, but is not stored in the β memory node Another case not fired In this case, the condition comparison in the inter node is performed. This is a cause of a malfunction not intended by the user, and when there are multiple internodes, it is necessary to identify which internode.

Here, the user is "A company-B company-C company".
Let's expect that Rule 1 will fire with the frame combination of. However, when the contents of the memory node are confirmed, the combination of "company A-company B" is stored in the β memory node 719, and company C is stored in the α memory node 717. However, "A company-B company-C company" is not stored in the terminal node 722, and the inter node 7
It is speculated that the matching in the 20 or 721 comparison failed.

Therefore, in order to identify which node the user is, the user selects "A company--" from the contents of the memory node 719.
The combination of "Company B" is selected, Company C is selected from the memory node 717, and the knowledge verification button 705 is instructed to start the knowledge verification.

In this case, regarding the condition that “the number of stores is the same” in the node 11, the value of the “number of stores” slot of company A does not match the value of the “number of stores” slot of company C. I understand. Therefore, the corresponding internode 721 and the corresponding conditional items and frame contents of the rule are highlighted. That is, the comparison part of the number of stores of the first conditional clause and the comparison part of the number of stores of the second conditional clause of rule 1 displayed in the rule display area 701, the store of company A displayed in the frame content display area 602 Several slots, the number of slots of the store of company C displayed in the frame content display area 603 is highlighted.

In this way, to find a defective part,
It can be performed by the algorithm shown in FIG. First, the information in the rule net information storage means 108 is searched,
All internodes that are performing the condition comparison using the two specified memory nodes are taken out (S90).
1).

Next, one of the memory nodes is temporarily saved, and the content of the memory node is limited to the designated frame or frame combination (for example, in the example shown above, the content of the β memory is temporarily set to A). Company-B company only) (S902).

Then, the contents of the other memory node are not changed, and the designated frame or frame combination is used as a change frame (S903), and the inference means 100 is activated for each searched internode to perform condition comparison ( S904, S905, S906).

As a result, if the condition comparison fails, the internode becomes a cause of trouble, so the internode is taken out (S907). Finally, it is necessary to restore the saved memory node information (S9).
08).

Further, when it is desired to find a plurality of defective points at the same time, the algorithm shown in FIG. 12 can be used. That is, even if a defective inter-node is found (S907), the processing is not ended there, but a loop may be made to perform the next condition comparison (S90a).

Also in this case, as in the case 1, since more defective parts can be known by one debugging operation, it can be expected to further contribute to the progress of the debugging work by the user.

[Second Embodiment] A second embodiment of the present invention will be described below with reference to FIGS. 13 to 17. (I) Absence condition The second embodiment according to the present invention is basically based on the first embodiment, but relates to a knowledge verification support system when a rule includes an "absence condition". Is. This absent condition is a condition that there is no certain frame that satisfies the rule.
Here, a condition theory including an absence condition is called an "absence condition clause".

Although such a condition is effective in improving the description ability of the rule and expanding the range that can be expressed by the knowledge system, special control is also required in the inference method, and correspondingly, It is necessary to consider providing knowledge verification support methods.

Specifically, the rule including the absent condition is
Let us explain based on FIG. 13 and FIG. FIG. 13 is a diagram showing rules used in the second embodiment according to the present invention. FIG. 14 shows the rule shown in FIG.
It is a schematic diagram expanded and shown in a network.

As shown in FIG. The condition described in (Condition) is the absolute condition clause. The mark represents the meaning of Absense, meaning that there is no frame that satisfies the condition. Therefore, the rule shown in FIG. 13 means "if the listed company on the First Section of the Tokyo Stock Exchange and the company that has exactly the same capital and number of stores does not exist on the second section of the Tokyo Stock Exchange, execute the following things." Become.

A rule including this absent conditional clause is Re
FIG. 14 shows the result when expanded to the te network. Here, the double triangle of the symbol in the figure is a node of the network that performs comparison in consideration of the absence condition, and is referred to as an “absence condition comparison node”. In this case,
When at least one of the condition of the node 25 and the condition of the node 26 is satisfied, it means that the absense condition is satisfied.

In the absense condition section, when the absense symbol is removed and considered, if there is a combination of frames satisfying the absense condition, the absense condition is not satisfied. Therefore, the frame or frame combination existing in the memory node of the conditional clause (hereinafter, referred to as “normal conditional clause” in FIG. 14) without the absence symbol is the memory node after the absolute condition comparison node. If it does not exist, it means that there is a frame that prevents the absense condition from being satisfied.

Therefore, in the conventional method, it is necessary to perform the visual confirmation work on all the frames which may be such a hindrance factor, which is much more troublesome than the visual confirmation work in the case of the first embodiment. Become. It can be said that the present embodiment is a very useful method for saving such trouble.

(II) Procedure of Knowledge Verification Support Method Regarding Rule Including Absence Condition and Its User Interface Next, based on the above, (I) the rule and the frame shown in FIG. 16 and FIG. 17, the procedure of the knowledge verification support method regarding the rule including the absense condition and its user interface will be described. FIG. 15 is a diagram showing a frame used in the second embodiment according to the present invention. FIG. 16 is a diagram showing a screen display example of the knowledge verification support system according to the knowledge verification support method of the second embodiment of the present invention. FIG. 17 is a flowchart showing an algorithm for finding a defect when inferred by a rule including an absent condition. Let us assume that the user is expecting that “Rule 2 holds true for“ Company A ””. In addition, the user also has nodes 1602 and 1603 of the rule net displayed in the network display area 1601 of FIG.
By looking at the double triangular symbol of
It can be recognized that the rule includes an absence condition.
Then, when the node 1602 and the node 1603 are selected with the mouse and the contents display button is designated, the corresponding condition portion of the rule display area is highlighted with high brightness. In this example, the comparison part regarding “capital” and “number of stores” in the first and second condition sections is highlighted.

Next, the contents of the memory node are confirmed from the network display area 1601. In this case, is stored in the memory node 1906 and the memory node 160
In FIG. 7, frames [“Company B”, “Company C”, “Company D”] that satisfy the conditions without the absent sign (!) Are stored. However, contrary to the user's expectation, the terminal node 1608 does not store a frame that causes a fire.

Therefore, it can be inferred that the comparison of this rule has failed in the comparison of the absolute condition comparison nodes 1602 and 1603.

Therefore, a frame including an inhibiting factor that inhibits the establishment of this absent condition is specified. Specifically, ["A stored in the memory node 1607]
"Company"] is selected, and a defect detection button 1609 is designated.

As a result, the column corresponding to Company A is highlighted with high brightness, and the knowledge verification means 109 is activated.
From the frames stored in the memory node 1607,
The column corresponding to the frame including the factor that inhibits the establishment of the absent condition is highlighted with high brightness. Then, the "capital" and "the number of stores", which are the slots of the frames that are compared and contrasted with each other in the frame display areas 1610 and 1611, are highlighted with high brightness, which causes a problem that the user's expectations cannot be met. Company B and Company B have the same capital and number of stores, which clearly indicates that this has occurred.

As described above, the defective portion relating to the rule including the absent conditional clause can be found by the algorithm shown in FIG. First, the frame or frame combination stored in the memory node of the normal condition clause which is desired to be checked by the user and is to be examined for the cause of the failure of the absense condition is retrieved, and the rule net information storage means 108 is searched based on the frame or frame combination, and the next memory is searched. Extract all the absolute condition nodes up to the node (S1
101).

Therefore, the frame or frame combination designated by the user is set as the changed frame or changed frame combination (S1102).

On the other hand, there is an α memory node on the side of the absent condition node as well, and stores a frame satisfying the condition when replaced with the normal condition. Save this (S110
3) Then, one of the frames is set in the memory node for absense (S1104). Then, the inference means 100 is activated according to the retrieved absolute condition node to perform condition comparison (S1105). Then, the contents of the absense α memory node for which the absense comparison has failed on the way are stored (S1109).

The steps from S1105 to S1105 are performed for all the absent condition nodes (S110).
8).

Then, the steps from S1104 to S1110 are performed over all the target changed frames.
Repeatedly (S1110).

Finally, it is necessary to restore the contents of the Absense α memory to the original contents (S1111).

As described above, in this way, S1
The frame stored in the memory node for absense stored in the step 109 is displayed as a frame that inhibits the establishment of the absense condition specified by the user.

[Possibility of Application of the Present Invention Other Than Rete Network] In the first and second embodiments, the rule is Ret.
We have explained the knowledge verification method when deploying to e-networks and making inferences based on it. However, even in networks that do not have β memory (treat networks) or even networks that do not have even α nodes, the rules that should fire and the It is possible to find the cause of the defect by performing local matching so as not to affect the entire matching process with the frame combination.

[0101]

According to the present invention, in debugging a knowledge processing system that describes knowledge by rules and frames, when a problem "a rule that should have fired does not fire" occurs for a rule that has not fired, , It is possible to provide a knowledge verification support method that allows quick investigation of the cause of the rule not firing, which is different from the user's expectation, and facilitates debugging work.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of a knowledge processing system having knowledge verification means.

FIG. 2 is a diagram showing rules used in the first embodiment according to the present invention.

FIG. 3 is a schematic diagram showing the rule shown in FIG. 2 developed in a Rete network.

FIG. 4 is a schematic diagram showing an expression on a memory cell of the Rete network.

FIG. 5 is a diagram showing a frame used in the first embodiment according to the present invention.

FIG. 6 is a general flowchart showing a procedure of the knowledge verification support method according to the present invention.

FIG. 7 is a diagram showing a screen display example of the knowledge verification support system according to the knowledge verification support method of the first embodiment of the present invention (part 1).

FIG. 8 is a diagram showing a screen display example of the knowledge verification support system according to the knowledge verification support method of the first embodiment of the present invention (No. 2).

FIG. 9 is a flowchart showing an algorithm for finding a defect in case 1 of searching an intra node (part 1).

FIG. 10 is a flowchart showing an algorithm for finding a defect in case 1 of searching an intra node (second).

FIG. 11 is a flowchart showing an algorithm for finding a defect in case 2 of searching an internode (part 1).

FIG. 12 is a flowchart showing an algorithm for finding a defect in case 2 of searching an internode (part 2).

FIG. 13 is a diagram showing rules used in the second embodiment according to the present invention.

FIG. 14 is a schematic diagram showing the rule shown in FIG. 13 expanded to a Rete network.

FIG. 15 is a diagram showing a frame used in the second embodiment according to the present invention.

FIG. 16 is a diagram showing a screen display example of the knowledge verification support system according to the knowledge verification support method of the second embodiment of the present invention.

FIG. 17 is a flowchart showing an algorithm for finding a defect when inferred by a rule including an absence condition.

[Explanation of symbols]

101 ... Frame storage means, 102 ... Rule net storage means, 103 ... Rule storage means, 104 ... Frame display means, 105 ... Rule net display means, 106 ... Rule display means, 107 ... Inference control means, 108 ... Rule net information storage Means, 109 ... Knowledge verification means, 100 ... Inference means, 110 ... Inference section, 111 ... Knowledge verification support section, 11
2 ... Input means.

Claims (4)

[Claims] 1. A knowledge processing system for performing inference by describing knowledge in terms of "facts" related to a target problem area by a frame and "judgment rules" that lead to another fact from the facts by rules, which comprises inference means. A means for performing knowledge verification, a means for storing the frame and the rule, a knowledge state display means, and an input means for specifying the frame and the rule subject to knowledge verification and giving a command to the system In the knowledge verification support method for a knowledge processing system, the state of a frame being inferred and the collation state of a rule are displayed on the knowledge state display means, and the frame and rule specified by the user by the input means are targeted for knowledge verification. , Finding the cause that the specified rule has not fired using the knowledge verification means, The knowledge verification support method of knowledge processing system, characterized in that the rule is displayed with an explicit condition part and the frame of the rule that caused the non-ignition and the value of the frame. 2. The rule is described as being unfulfilled when there is at least one frame in which some of the conditions are not satisfied, and this rule is subject to knowledge verification. In this case, a frame in which a part of the conditions is not satisfied is found by using the knowledge verification means as a cause of unfired, and the frame is explicitly displayed on the knowledge state display means. A knowledge verification support method for a knowledge processing system according to claim 1. 3. A comparison node that expands a rule and performs a condition comparison process corresponding to each condition of a judgment rule, and a memory node that performs a storage process of facts partially satisfying the condition of the judgment rule. When performing inference using the matching processing network, it is possible to display the structure of the matching processing network as a network structure diagram on the knowledge state display means, and display the matching state of the inference result on the network structure diagram. A knowledge verification support method for a knowledge processing system according to any one of claims 1 and 2. 4. The frame stored in the memory node is displayed on the network structure diagram, the frame is specified by the input means, the knowledge verification is performed as a target of the knowledge verification, and the rule of which the verification fails is verified. 4. The knowledge verification support method for a knowledge processing system according to claim 3, wherein the comparison node corresponding to the condition section is clearly indicated on the network structure diagram.