Inference method and inference system

• US 5,899,985 A
• Filed: 09/05/1995
• Issued: 05/04/1999
• Est. Priority Date: 09/05/1994
• Status: Expired due to Fees

• Current Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Original Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Inventors: Tanaka, Toshikazu
• Primary Examiner(s): Downs, Robert W.
• Assistant Examiner(s): Shah, Sanjiv
• Application Number:
  US08/523,617
• Time in Patent Office: 1,337 Days
• Field of Search: 706/45, 706/54, 706/46, 706/53
• US Class Current: 706/45
• CPC Class Codes: G06N 5/04 Inference or reasoning models
• CPC Class Codes: G06N 5/046 Forward inferencing; Production systems

Abstract

It is an object of this invention to provide a high-accuracy inference method and an inference system. The input reads fact data into this system. Fact data is stored in a work area, and the inference is started. The inference cycle of the inference is as follows. First, the rule extraction processing module compares rules with data to extract an optimum rule. Then, the case extraction processing module extracts from the case storing a case which matches the extracted rule. Then, the selection processing module compares the condition of the extracted rule with that of the extracted case, and selects one of them which is more similar to the fact data. The execution processing module executes the procedure of the selected rule or case. The inference cycle is repeated unless the executed procedure stops processing.

Claims

1. An inference method which performs inference using

a set of rules, each consisting of a conditional statement describing generally a condition necessary for inference and a conclusion statement describing a conclusion corresponding to the condition;

and

a set of cases, each consisting of a conditional statement describing a condition representing a case and a result statement describing a result corresponding to the condition, said inference method comprising;

a rule extracting process for extracting said rule having a condition which matches given input data;

a case extracting process for extracting said case corresponding to an extracted rule;

an evaluating process for evaluating the similarity of said rule and said case to said input data;

a selecting process for selecting either said rule or said case which has a higher similarity to said input data; and

an executing process for executing inference based on the conclusion statement of said selected rule or on the result statement of said selected case.

3. An inference method as claimed in claim 1 or 2 wherein said rule extraction process comprising:

a matching process for comparing the condition described in the conditional statement of each rule stored in said rule storing process with data stored in said work area and for extracting a rule matching each data; and

a conflict resolution process for selecting one rule, by using a specified method, when a plurality of rules are extracted by said matching process.

4. An inference method as claimed in claim 1 or 2, further comprising:

a similarity evaluation knowledge storing process for storing the weights of the attributes constituting said conditions and similarity evaluation knowledge such as a similarity degree function for evaluating similarity between said condition and said fact data, and wherein a similarity degree used in comparison in said selection process is a value satisfying ##EQU9## where, P is said condition, Q is said fact data, sim is a similarity between condition P and fact data Q, n is the number of attributes for condition P, wt_i (i=1 to n) is a weight stored in said similarity evaluation knowledge storing process, and sim_i ( ) (i=1 to n) is a similarity degree function stored in said similarity evaluation knowledge storing process.

5. An inference method as claimed in claim 4, further comprising:

an inference history storing process for storing inference processing history for each cycle in said inference process,

a re-executing process for re-executing inference processing, after inference processing in said inference process is finished, in said inference process according to inference processing history stored in said inference history storing process,

an evaluation input process for accepting expert'"'"'s evaluation on the inference process and inference result of inference processing performed in said re-executing process, and

a case adding process for classifying said inference result into a correct case (positive case) and an incorrect case (negative case) according to the evaluation of an expert entered via said evaluation input process and for adding to said case storing process this evaluation result as well as a case whose conditional statement is said fact data and whose result statement is said inference result,

wherein said inference process comprises a history storing process for storing the history of inference processing in each cycle into said inference history storing process.

6. An inference method as claimed in claim 5, further comprising a similarity evaluation knowledge adjusting process for adding and changing, with the use of a specified method, similarity evaluation knowledge stored in said similarity evaluation knowledge storing process according to expert'"'"'s evaluation entered via said evaluation input process.

7. An inference method as claimed in claim 5, further comprising rule adding process for creating a new rule from each case stored in said case storing process and for adding the created rule to said rule storing process while referencing similarity evaluation knowledge stored in said similarity evaluation knowledge storing process.

2. An inference method comprising:

a rule storing process for storing a plurality of rules, each consisting of a conditional statement describing a condition required for inference and a conclusion statement describing a conclusion procedure judged by an expert; and

a case storing process for storing a plurality of cases, each consisting of a conditional statement describing a condition used in an actual case and a result statement describing a result procedure executed under the condition;

using a work area for temporarily storing problem data entered via input process and hypothetical data generated during inference process;

further comprising;

an input process for accepting fact data entered by a user and for storing it into said work area; and

an inference process for repeating a cycle until a processing stop procedure is executed, in which a rule stored in said rule storing process or a case stored in said case storing process, each corresponding to data stored in said work area, is selected and then the conclusion statement of as elected rule or the result statement of a selected case is executed,

wherein one cycle of said inference process comprising;

a rule extraction process for extracting an optimum rule by comparing data stored in said work area with a condition described in the conditional statement of each rule stored in said rule storing process;

a case extraction process for extracting a case having a conditional statement describing at least partially matching with a condition described in the conditional statement of a rule extracted by said rule extraction process;

a selection process for comparing the condition described in the conditional statement of the rule extracted from said rule extraction process with the condition described in the conditional statement of the case extracted by said case extraction process and for selecting either the rule or the case which is more similar to fact data entered said input process; and

an execution process for executing the procedure in the conclusion statement of the rule or in the result statement of the case selected by said selection process.

8. An inference system which performs inference using

a set of rules, each consisting of a conditional statement describing generally a condition necessary for inference and a conclusion statement describing a conclusion corresponding to the condition;

and

a set of cases, each consisting of a conditional statement describing a condition representing a case and a result statement describing a result corresponding to the condition, said inference system comprising;

a rule extracting means for extracting said rule having a condition which matches given input data;

a case extracting means for extracting said case corresponding to an extracted rule;

an evaluating means for evaluating the similarity of said rule and said case to said input data;

a selecting means for selecting either said rule or said case which has a higher similarity to said input data; and

an executing means for executing inference based on the conclusion statement of said selected rule or on the result statement of said selected case.

10. An inference system as claimed in claim 8 or 9 wherein said rule extraction processing module comprising:

a matching processing module for comparing the condition described in the conditional statement of each rule stored in said rule storing means with data stored in said work area and for extracting a rule matching each data; and

a conflict resolution processing module for selecting one rule, by using a specified method, when a plurality of rules are extracted by said matching processing module.

11. An inference system as claimed in claim 8 or 9, further comprising:

a similarity evaluation knowledge storing means for storing the weights of the attributes constituting said conditions and similarity evaluation knowledge such as a similarity degree function for evaluating similarity between said condition and said fact data, and wherein a similarity degree used in comparison in said selection processing module is a value satisfying ##EQU10## where, P is said condition, Q is said fact data, sim is a similarity between condition P and fact data Q, n is the number of attributes for condition P, wt_i (i=1 to n) is a weight stored in said similarity evaluation knowledge storing means, and sim_i (i=1 to n) is a similarity degree function stored in said similarity evaluation knowledge storing means.

12. An inference system as claimed in claim 11, further comprising:

an inference history storing means for storing inference processing history for each cycle in said inference means,

a re-executing means for re-executing inference processing, after inference processing in said inference means is finished, in said inference means according to inference processing history stored in said inference history storing means,

an evaluation input means for accepting expert'"'"'s evaluation on the inference process and inference result of inference processing performed in said re-executing means, and

a case adding means for classifying said inference result into a correct case (positive case) and an incorrect case (negative case) according to the evaluation of an expert entered via said evaluation input means and for adding to said case storing means this evaluation result as well as a case whose conditional statement is said fact data and whose result statement is said inference result,

wherein said inference means comprises a history storing processing module for storing the history of inference processing in each cycle into said inference history storing means.

13. An inference system as claimed in claim 12, further comprising a similarity evaluation knowledge adjusting means for adding and changing, with the use of a specified method, similarity evaluation knowledge stored in said similarity evaluation knowledge storing means according to expert'"'"'s evaluation entered via said evaluation input means.

14. An inference system as claimed in claim 12, further comprising a rule adding means for creating a new rule from each case stored in said case storing means and for adding the created rule to said rule storing means while referencing similarity evaluation knowledge stored in said similarity evaluation knowledge storing means.

9. An inference system comprising:

a rule storing means for storing a plurality of rules, each consisting of a conditional statement describing a condition required for inference and a conclusion statement describing a conclusion procedure judged by an expert;

a case storing means for storing a plurality of cases, each consisting of a conditional statement describing a condition used in an actual case and a result statement describing a result procedure executed under the condition;

a work area for temporarily storing problem data entered via input means and hypothetical data generated during inference process;

an input means for accepting fact data entered by a user and for storing it into said work area; and

an inference means for repeating a cycle until a processing stop procedure is executed, in which a rule stored in said rule storing means or a case stored in said case storing means, each corresponding to data stored in said work area, is selected and then the conclusion statement of a selected rule or the result statement of a selected case is executed,

wherein one cycle of said inference means comprising;

a rule extraction processing module for extracting an optimum rule by comparing data stored in said work area with a condition described in the conditional statement of each rule stored in said rule storing means;

a case extraction processing module for extracting a case having a conditional statement describing at least partially matching with a condition described in the conditional statement of a rule extracted by said rule extraction processing module;

a selection processing module for comparing the condition described in the conditional statement of the rule extracted from said rule extraction processing module with the condition described in the conditional statement of the case extracted by said case extraction processing module and for selecting either the rule or the case which is more similar to fact data entered said input means; and

an execution processing module for executing the procedure in the conclusion statement of the rule or in the result statement of the case selected by said selection processing module.