Method, program product, and apparatus for generating analysis model

• US 7,630,951 B2
• Filed: 01/09/2007
• Issued: 12/08/2009
• Est. Priority Date: 03/17/2006
• Status: Expired due to Fees

• Current Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Original Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Inventors: Kobayashi, Hideki, Murakami, Tomoko
• Primary Examiner(s): Holmes; Michael B
• Application Number:
  US11/650,934
• Publication Number:
  US 20070233632A1
• Time in Patent Office: 1,064 Days
• Field of Search: 706/52
• US Class Current: 706/52
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks

Abstract

The present invention provides an analysis model generating method for generating an analysis model expressed using a Bayesian network. A analysis model generating method includes selecting a child variable to be added to the analysis model; selecting a plurality of variable candidates to be added to the analysis model as parent variables in causal relationships with the child variable; judging whether each of values indicating the causal relationships between the selected variable candidates and the child variable is high or low, based on the causal relationships between the variable candidates and the child variable; generating an aggregated variable different from the variable candidates by aggregating a plurality of low causal variable candidates; and putting high causal variable candidate and the aggregated variable into the analysis model as the parent variables to the child variable.

Claims

1. An analysis model generating method for generating an analysis model in which causal relationships between variables are expressed using a Bayesian network, the method comprising:

selecting a child variable to be added to the analysis model;

selecting a plurality of variable candidates to be added to the analysis model as parent variables in causal relationships with the child variable;

judging, using a processor, whether each of values indicating the causal relationships between the selected variable candidates and the child variable are high causal variable candidates or low causal variable candidates, based on the causal relationships between the variable candidates and the child variable;

generating an aggregated variable different from the variable candidates by aggregating a plurality of low causal variable candidates, each of the low causal variable candidates being judged to have a low value in indicating the causal relationship;

generating the analysis model by putting a high causal variable candidate and the aggregated variable into the analysis model as the parent variables to the child variable, the high causal variable candidate being judged to have a high value in indicating the causal relationship; and

evaluating target data by using the analysis model.

2. The analysis model generating method according to claim 1, wherein

the variable candidates are selected based on information related to variables included in target data that is an analysis target to be analyzed by using the analysis model.

3. The analysis model generating method according to claim 2, wherein

the causal relationship between the variable candidate and the child variable are evaluated based on the information related to the variables included in the target data, and

the variable candidates are selected based on the evaluated causal relationships.

4. The analysis model generating method according to claim 1, wherein the variable candidates are selected based on a specification by a user.

5. The analysis model generating method according to claim 1, wherein

the aggregated variable is generated by aggregating the low causal variable candidate and the high causal variable candidate.

6. The analysis model generating method according to claim 1, wherein

each of all variables to be included in the analysis model is selected as the child variable, and

for each of all the child variables, selecting the variable candidates, generating the parent variables, and determining the parent variables are recursively repeated.

7. The analysis model generating method according to claim 6, wherein

the causal relationship between the child variable and the parent variable is changed into causal relationship between the aggregated variable and the parent variable, when parent variable to a child variable is determined, the child variable being one of the variable candidates aggregated into the aggregated variable.

8. A computer program product having a computer-readable medium including programmed instructions for performing an analysis model generating process to generate an analysis model in which causal relationships between variables are expressed using a Bayesian network, wherein the instructions, when executed by a computer, cause the computer to perform:

selecting a child variable to be added to the analysis model;

selecting a plurality of variable candidates to be added to the analysis model as parent variables in causal relationships with the child variable;

judging, using a processor, whether each of values indicating the causal relationships between the selected variable candidates and the child variable are high causal variable candidates or low causal variable candidates, based on the causal relationships between the variable candidates and the child variable;

generating an aggregated variable different from the variable candidates by aggregating a plurality of low causal variable candidates, each of the low causal variable candidates being judged to have a low value in indicating the causal relationship;

generating the analysis model by putting a high causal variable candidates and the aggregated variable into the analysis model as the parent variable to the child variable, the high causal variable candidate being judged to have a high value in indicating the causal relationship; and

evaluating target data by using the analysis model.

9. An analysis model-generating apparatus for generating an analysis model in which causal relationships between variables are expressed using a Bayesian network, the apparatus comprising:

a processor;

a child variable selecting unit that selects a child variable to be added to the analysis model;

a variable candidate selecting unit that selects a plurality of variable candidates to be added to the analysis model as parent variables in causal relationships with the child variable;

a causal relationship judging unit that judges whether each of values indicating the causal relationships between the selected variable candidates and the child variable are high causal variable candidates or low causal variable candidates, based on the causal relationships between the variable candidates and the child variable;

an aggregated variable generating unit that generates an aggregated variable different from the variable candidates by aggregating a plurality of low causal variable candidates, each of the low causal variable candidates being judged to have a low value in indicating the causal relationship; and

a constructing unit that generates the analysis model by putting a high causal variable candidate and the aggregated variable into the analysis model as the parent variables to the child variable, the high causal variable candidate being judged to have a high value in indicating the causal relationship; and

an evaluating unit that evaluates target data by using the analysis model.