Reasoning method based on similarity of cases

• US 6,782,376 B2
• Filed: 01/08/2001
• Issued: 08/24/2004
• Est. Priority Date: 05/18/2000
• Status: Expired due to Fees

• Current Assignee: Hitachi, Ltd.
• Original Assignee: Hitachi, Ltd.
• Inventors: Ito, Yukiyasu, Sato, Yoshinori, Morita, Toyohisa
• Primary Examiner(s): Starks, Jr., Wilbert L.
• Assistant Examiner(s): Booker, Kelvin
• Application Number:
  US09/755,138
• Publication Number:
  US 20010054033A1
• Time in Patent Office: 1,324 Days
• Field of Search: 706/47, 706/46, 706/45, 706/21, 706/19, 706/2, 706/6, 706/8, 706/48
• US Class Current: 706/47
• CPC Class Codes: G06N 5/04 Inference or reasoning models

Abstract

A reasoning method for determining similar cases by executing a case-similarity calculation step and utilizing a distribution of values of fields for calculation of reasoning results of cases having high similarities. Since the similar cases are determined taking a peripheral distribution of a new case into consideration, suitable similar cases can be selected for reasoning even when the case distribution is not uniform.

Claims

1. A method for calculating a confidence degree as a degree of certainty in reasoning with use of similarities between cases of a record type including at least two fields, the method being implemented by a computer system and the cases being stored in a database, the method comprising the steps of:

setting at least one input field for use in calculation of the similarities between the cases and at least one output field to be an object of the calculation of a reasoning result;

inputting parameters relating to the similarity calculation including an allowable value of a reasoning error;

inputting a new case having at least one output field having an unknown value;

calculating similarities between said new case and previous cases already stored for each of the previous cases;

generating a distribution of the output fields by referring to values of the output fields of the previous cases in a decreasing order of said similarities;

comparing said distribution with said allowable value to calculate a confidence degree in the reasoning;

calculating a reasoning result of the output field value of the new case with use of the output field values of the previous cases and said calculated similarities; and

outputting said reasoning result and said confidence degree.

2. A reasoning method as set forth in claim 1, wherein the parameter relating to said similarity calculation has a value indicative of a fluctuation in an allowable reasoning result and has a lower limit value of the number of similar cases in the previous cases for use in calculation of one reasoning result, and

the confidence degree is a number of previous cases which output field values that are within the fluctuation in an allowable reasoning result.

3. A reasoning method as set forth in claim 1, wherein the parameter relating to said similarity calculation has a value indicative of a fluctuation in an allowable reasoning result, a number of similar cases in the previous cases for use in calculation of one reasoning result, and a lower limit value of a ratio between the number of the similar cases not exceeding the fluctuation of said allowable reasoning result and the number of the similar cases for use in the one reasoning result, and

the confidence degree is the ratio between the number of similar cases not exceeding the fluctuation of said allowable reasoning result and the number of the similar cases for use in the one reasoning result.

4. A reasoning method as set forth in claim 1, wherein the parameter relating to said similarity calculation has a lower limit value of the number of similar cases of the previous cases for use in one reasoning result and has a value for restriction of the distribution of the similar cases.

5. A method for calculating a confidence degree as a degree of certainty in reasoning with use of similarities between cases of a record type including at least two fields, the method being implemented by a computer system and the cases being stored in a database, the method comprising the steps of:

(a) setting at least one input field for use in calculation of the similarities between the cases and at least one output field to be an object of the calculation of a reasoning result;

(b) inputting parameters relating to the similarity calculation including an allowable value of a reasoning error and an allowable confidence degree, the confidence degree being a number of previous cases satisfying a condition of the reasoning error;

(c) inputting a new case having at least one output field having an unknown value;

(d) calculating similarities between said new case and previous cases already stored for each of the previous cases;

(e) generating a distribution of the output fields by referring to values of the output fields of the previous cases in a decreasing order of said similarities;

(f) comparing said distribution with said allowable value to calculate the confidence degree in the reasoning;

(g) calculating a reasoning result of the output field value of the new case with use of the output field values of the previous cases and said calculated similarities; and

(h) judging whether or not to repeat said steps (b) to (g) on the basis of output results of said reasoning result and said confidence degree.

6. A reasoning method for finding a reasoning result on the basis of similarities between cases of a record type including at least two fields, the method being implemented by a computer system and the cases being stored in a database, the method comprising the steps of:

setting at least one input field having a value already set and at least one output field as an object to be reasoned;

inputting a reasoning parameter to find a reasoning confidence degree;

inputting a new case having at least one output field having an unknown value;

calculating similarities of previous cases for said new case on the basis of the value of said input field;

generating a distribution of values of the output fields of the previous cases selected on the basis of said similarities;

calculating a reasoning confidence degree on the basis of said distribution and said reasoning parameter;

calculating a reasoning result for the value of the output field of the new case on the basis of the values of the output fields of said selected previous cases and said similarities; and

outputting said reasoning result and said confidence degree.

7. A reasoning system for finding a reasoning result on the basis of similarities between cases of a record type including at least two fields, the cases being stored in a database, the system comprising:

means for setting at least one input field having a value already set and at least one output field as an object to be reasoned;

means for inputting a reasoning parameter to find a reasoning confidence degree;

means for inputting a new case having at least one output field having an unknown value;

means for calculating similarities of previous cases for said new case on the basis of the value of said input field;

means for generating a distribution of values of the output fields of the previous cases selected on the basis of said similarities;

means for calculating a reasoning confidence degree on the basis of said distribution and said reasoning parameter;

means for calculating a reasoning result for the value of the output field of the new case on the basis of the values of the output fields of said selected previous cases and said similarities; and

means for outputting said reasoning result and said confidence degree.

8. A reasoning method for finding a reasoning result on the basis of similarities between cases of a record type including at least two fields, the method being implemented by a computer system and the cases being stored in a database, the method comprising the steps of:

calculating similarities between a new case having at least one output field having an unknown value and previous cases based on input fields already set with values;

calculating a reasoning confidence degree on the basis of a distribution of values of output fields of the previous cases selected based on said similarities and a previously-input reference value;

calculating a reasoning result of the value of the output field of the new case on the basis of the values of the output fields of said selected previous cases and said similarities; and

outputting said reasoning result and said confidence degree.

9. A reasoning method for finding a reasoning result on the basis of similarities between cases of a record type including at least two fields, the method being implemented by a computer system and the cases being stored in a database, the method comprising the steps of:

calculating similarities between a new case having at least one output field having an unknown value and previous cases having all fields already set with values using input fields already set with values of the new and previous cases;

selecting some of the previous cases in accordance with the calculated similarities; and

calculating a reasoning result of the value of the output field of the new case on the basis of the values of the output fields of the selected previous cases and said calculated similarities.