Systems, methods, and computer program products for generating a query specific Bayesian network

• US 10,504,030 B2
• Filed: 07/25/2015
• Issued: 12/10/2019
• Est. Priority Date: 07/25/2015
• Status: Active Grant

• Current Assignee: The Boeing Co.
• Original Assignee: The Boeing Co.
• Inventors: Kipersztok, Oscar, Nodelman, Uri
• Primary Examiner(s): Chang, Li Wu
• Application Number:
  US14/809,179
• Publication Number:
  US 20170024652A1
• Time in Patent Office: 1,599 Days
• Field of Search: None
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks

Abstract

Provided are improved systems, methods, and computer programs to facilitate predictive accuracy for strategic decision support using a query specific Bayesian network. An unconstrained domain model is defined by domain concepts and causal relationships between the domain concepts. Each causal relationship includes a value for the weight of causal belief for the causal relationship. In response to a query, the unconstrained domain model is transformed into a query specific Bayesian network for the domain model by identifying one or more cycles in the unconstrained domain model, eliminating the one or more cycles from the unconstrained domain model; identifying a sub-graph of the unconstrained domain model that is relevant to a query and creating one or more conditional probability tables that comprise the query specific Bayesian network.

Claims

1. A method of a decision support system, comprising:

providing an unconstrained domain model defined by domain concepts and causal relationships between the domain concepts, wherein each causal relationship includes a value for the weight of causal belief for the causal relationship;

transforming, using processing circuitry and in response to a query, the unconstrained domain model into a query specific Bayesian network for the domain model by identifying one or more cycles in the unconstrained domain model;

eliminating the one or more cycles from the unconstrained domain model;

identifying a sub-graph of the unconstrained domain model that is relevant to the query and creating one or more conditional probability tables that comprise the query specific Bayesian network,

wherein eliminating the one or more cycles comprises identifying the causal relationships between the domain concepts that are weak relative to other causal relationships in the unconstrained domain model as a result of having a weight with an absolute value that fails to satisfy a predefined threshold that is based upon a representation of the weights of a plurality of the causal relationships in the unconstrained domain model, and removing the causal relationships between the domain concepts that are identified to be weak; and

utilizing the query specific Bayesian network to analyze textual data by performing textual and reasoning processing and to provide output based upon probabilistic predictions based thereupon.

2. The method of claim 1, wherein identifying the sub-graph of the unconstrained domain model comprises pruning domain concepts that are irrelevant to the query.

3. The method of claim 1 wherein identifying the sub-graph of the unconstrained domain model comprises identifying the domain concepts that are relevant to the query.

4. The method of claim 1, wherein identifying one or more cycles in the unconstrained domain model comprises identifying all cycles in the unconstrained domain model, and wherein eliminating the one or more cycles from the unconstrained domain model comprises eliminating all cycles from the unconstrained domain model.

5. The method of claim 1, wherein creating one or more conditional probability tables comprises creating a conditional probability table defining the causal relationship from a parent domain concept to a child domain concept by determining a probability of occurrence for a respective transition from each state of the parent domain concept to each state of the child domain concept.

6. The method of claim 1, wherein creating one or more conditional probability tables comprises creating a conditional probability table defining the causal relationship from a plurality of parent domain concepts to a child domain concept by determining a probability of occurrence for a respective transition from each combination of states of the parent domain concepts to each state of the child domain concept.

18. The method of claim 1 wherein the unconstrained domain model comprises child domain concepts and parent domain concepts, and wherein, in an instance in which a child domain concept has a plurality of parent domain concepts, defining a conditional intensity matrix for each child-parent relationship pairing and amalgamating the conditional intensity matrices by matrix addition.

7. A decision support system, comprising:

a memory configured to provide an unconstrained domain model defined by domain concepts and causal relationships between the domain concepts, wherein each causal relationship includes a value for the weight of causal belief for the causal relationship; and

processing circuitry configured to transform, in response to a query, the unconstrained domain model into a query specific Bayesian network for the domain model by identifying one or more cycles in the unconstrained domain model;

eliminating the one or more cycles from the unconstrained domain model;

identifying a sub-graph of the unconstrained domain model that is relevant to the query and creating one or more conditional probability tables that comprise the query specific Bayesian network,

wherein eliminating the one or more cycles comprises identifying the causal relationships between the domain concepts that are weak relative to other causal relationships in the unconstrained domain model as a result of having a weight with an absolute value that fails to satisfy a predefined threshold that is based upon a representation of the weights of a plurality of the causal relationships in the unconstrained domain model, and removing the causal relationships between the domain concepts that are identified to be weak,

wherein the processing circuitry is configured to utilize the query specific Bayesian network to analyze textual data by performing textual and reasoning processing and to provide output based upon probabilistic predictions based thereupon.

8. The decision support system of claim 7, wherein the processing circuitry is configured to identify the sub-graph of the unconstrained domain model by pruning domain concepts that are irrelevant to the query.

9. The decision support system of claim 7 wherein the processing circuitry is configured to identify the sub-graph of the unconstrained domain model by identifying the domain concepts that are relevant to the query.

10. The decision support system of claim 7, wherein the processing circuitry is configured to identify one or more cycles in the unconstrained domain model by identifying all cycles in the unconstrained domain model, and wherein the processing circuitry is configured to eliminate the one or more cycles from the unconstrained domain model by eliminating all cycles from the unconstrained domain model.

11. The decision support system of claim 7, wherein the processing circuitry is configured to create one or more conditional probability tables by creating a conditional probability table defining the causal relationship from a parent domain concept to a child domain concept by determining a probability of occurrence for a respective transition from each state of the parent domain concept to each state of the child domain concept.

12. The decision support system of claim 7, wherein the processing circuitry is configured to create one or more conditional probability tables by creating a conditional probability table defining the causal relationship from a plurality of parent domain concepts to a child domain concept by determining a probability of occurrence for a respective transition from each combination of states of the parent domain concepts to each state of the child domain concept.

19. The system of claim 7 wherein the unconstrained domain model comprises child domain concepts and parent domain concepts, and wherein, in an instance in which a child domain concept has a plurality of parent domain concepts, the processing circuitry is further configured to define a conditional intensity matrix for each child-parent relationship pairing and amalgamate the conditional intensity matrices by matrix addition.

13. A computer program product of a decision support system comprising a non-transitory computer-useable medium having control logic stored therein, the control logic comprising:

a first code configured to providing an unconstrained domain model defined by domain concepts and causal relationships between the domain concepts, wherein each causal relationship includes a value for the weight of causal belief for the causal relationship; and

a second code configured to transform, in response to a query, the unconstrained domain model into a query specific Bayesian network for the domain model by identifying one or more cycles in the unconstrained domain model;

eliminating the one or more cycles from the unconstrained domain model;

identifying a sub-graph of the unconstrained domain model that is relevant to the query and creating one or more conditional probability tables that comprise the query specific Bayesian network,

wherein eliminating the one or more cycles comprises identifying the causal relationships between the domain concepts that are weak relative to other causal relationships in the unconstrained domain model as a result of having a weight with an absolute value that fails to satisfy a predefined threshold that is based upon a representation of the weights of a plurality of the causal relationships in the unconstrained domain model, and removing the causal relationships between the domain concepts that are identified to be weak,

wherein the query specific Bayesian network is configured to analyze textual data by performing textual and reasoning processing and to provide output based upon probabilistic predictions based thereupon.

14. The computer program product of claim 13, wherein the second code is configured to identify the sub-graph of the unconstrained domain model by pruning domain concepts that are irrelevant to the query.

15. The computer program product of claim 13, wherein the second code is configured to identify one or more cycles in the unconstrained domain model by identifying all cycles in the unconstrained domain model, and wherein the second code is configured to eliminate the one or more cycles from the unconstrained domain model by eliminating all cycles from the unconstrained domain model.

16. The computer program product of claim 13, wherein the second code is configured to create one or more conditional probability tables by creating a conditional probability table defining the causal relationship from a parent domain concept to a child domain concept by determining a probability of occurrence for a respective transition from each state of the parent domain concept to each state of the child domain concept.

17. The computer program product of claim 13, wherein the second code is configured to create one or more conditional probability tables by creating a conditional probability table defining the causal relationship from a plurality of parent domain concepts to a child domain concept by determining a probability of occurrence for a respective transition from each combination of states of the parent domain concepts to each state of the child domain concept.

20. The computer program product of claim 13 wherein the unconstrained domain model comprises child domain concepts and parent domain concepts, and wherein, in an instance in which a child domain concept has a plurality of parent domain concepts, the second code is further configured to define a conditional intensity matrix for each child-parent relationship pairing and amalgamate the conditional intensity matrices by matrix addition.