Interactive method to reduce the amount of tradeoff information required from decision makers in multi-attribute decision making under uncertainty

• US 10,366,331 B2
• Filed: 01/27/2016
• Issued: 07/30/2019
• Est. Priority Date: 03/15/2013
• Status: Active Grant

• Current Assignee: University College Cork (Higher Education Authority), International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Deleris, Lea A., Marinescu, Radu, Razak, Abdul, Wilson, Peter Nicholas
• Primary Examiner(s): Chang, Li Wu
• Application Number:
  US15/008,034
• Publication Number:
  US 20160155053A1
• Time in Patent Office: 1,280 Days
• Field of Search: None
• CPC Class Codes: G06N 5/00 Computing arrangements using knowledge-based models
• CPC Class Codes: G06N 5/02 Knowledge representation; Symbolic representation
• CPC Class Codes: G06N 5/022 Knowledge engineering; Knowledge acquisition
• CPC Class Codes: G06N 5/04 Inference or reasoning models

Abstract

There is provided a method, a system and a computer program product for supporting a decision making process. The system receives a decision model from a decision maker, the decision model used for determining a solution to a decision problem based on attributes and uncertainties of the decision problem. The decision problem includes information about a plurality of outcome vectors that represent all possible outcomes and the uncertainties associated with the decision problem. The system determines whether the received decision model can be solved without receiving any preference information from the decision maker. The system receives partially specified preference information from the decision maker if the received decision model cannot be solved without any preference information. The system solves the decision model with the partially specified preference information. The system recommends, based on the solution, one or more decisions to the decision maker.

Claims

1. A computer-implemented method for supporting a multi-attribute healthcare decision problem requiring a decision among possible treatment options, the method comprising:

receiving, from a user, via an input to a computer system, a user'"'"'s indication of a maximum number of decision strategies from which the user is able to choose a strategy that solves the healthcare decision problem;

receiving, from the user, via the input to the computer system, a specification of an upper bound pre-determined time period within which a probabilistic decision tree model is run by a processor to solve the healthcare decision problem;

the computer system using a processor for running said probabilistic decision tree model based on the received inputs to minimize a number of comparison queries to be asked to the user, a comparison query requesting which outcome vector the user prefers between two outcome vectors representing possible outcomes of the healthcare decision problem, an outcome vector represented by multiple attributes associated with a treatment option;

using the processor of the computer system to elicit from a decision maker, based on the minimized number of comparison queries, a minimal amount of preference information with respect to the multiple attributes, wherein said eliciting a minimal amount of preference information comprises;

identifying, using the processor, N pairs of outcome vectors to present to the decision maker for preference assessment, said N pairs of outcome vectors identified by;

storing, in a memory unit associated with said processor, a first set of input outcome vectors of the decision problem;

allocating, in said memory unit, a memory structure defining a cone for storing a second set of N pairs of outcome vectors presented to the decision maker, said pairs of outcome vectors enumerated as a set (u, v);

estimating a decision maker preference of an outcome vector for a pair of outcome vectors, wherein if estimated that the decision maker prefers an outcome vector u over an outcome vector v, setting a score of a pair of outcome vectors (u, v) to be a number of undominated strategies obtained by solving the decision problem under an assumption that the outcome vector u is preferred to the outcome vector v;

otherwise, if estimated that the decision maker prefers an outcome vector v over an outcome vector u, setting a score of a pair of outcome vectors (v, u) to be a number of undominated strategies obtained by solving the decision problem under an assumption that the outcome vector v is preferred to the outcome vector u;

defining candidate pairs of outcome vectors as an ascending order of all the pairs of outcome vectors according to their corresponding scores;

for each candidate pair of the ordered outcome vector (u, v), confirming whether the decision maker prefers the outcome vector u over the outcome vector v, and in response to said confirming, adding the outcome vector pair (u, v) to the cone;

solving, using the processor, the decision problem with the current outcome vectors (u, v) stored in the cone;

evaluating whether the solution satisfies the decision maker;

if the decision maker does not satisfy the solution, generating outcome vectors u and v on a boundary of the cone by generating a linear combination of vectors already currently in the cone;

if the decision maker confirms that the outcome vector u is preferred over the outcome vector v, adding the pair of outcome vector (u, v) to the cone, otherwise, if an outcome vector v is preferred over the outcome vector u, adding the pair of outcome vector (v, u) to the cone; and

repeating the solving of the decision problem with the current cone, the evaluating, the generating, and the adding the pair of outcome vector (u, v) or outcome vector (v, u) to the cone until the solution satisfies the decision maker;

running, by the processor, the decision model to solve, based on the elicited minimal amount of preference information with respect to the multiple attributes, the decision problem within the pre-determined time period;

using a processor of the computer system to output, as the solution of the decision problem, a set of recommended actions;

wherein said decision maker makes a decision based on the set of recommended actions that avoids a full elicitation of preferences over all the multiple attributes.

2. The method according to claim 1, wherein the set of recommended actions are manageable by the decision maker so that the set of recommended actions satisfies a rule or is chosen by the decision maker.

3. The method according to claim 1 wherein prior to eliciting, implementing the decision model for determining a solution to a decision problem based on said multiple attributes and uncertainties of the decision problem, the decision problem including information about a plurality of outcome vectors that represent all possible outcomes and the uncertainties associated with the decision problem, the decision model precluding any preference information of the attributes and the outcome vectors.

4. The method according to claim 3, further comprising:

determining whether the decision model can be solved without receiving any preference information from the decision maker.

5. The method according to claim 1, further comprising:

presenting N elicitation questions, based on the identified N pairs of outcome vectors, to the decision maker, an elicitation question requesting of the decision maker which outcome vector the decision maker prefers between two outcome vectors; and

providing, by the decision maker, answers to the N elicitation questions as representing the minimal amount of preference information.

6. A computer system for supporting a multi-attribute healthcare decision problem requiring a decision among possible treatment options, the method comprising:

a memory device;

a processor coupled to the memory device, wherein the processor is configured to;

receive, from a user, via an input to a computer system, a user'"'"'s indication of a maximum number of decision strategies from which the user is able to choose a strategy that solves the healthcare decision problem;

receive from the user, via the input to the computer system, a specification of an upper bound pre-determined amount of time within which a probabilistic decision tree model is run by a processor to solve the healthcare decision problem;

run the probabilistic decision tree model based on the received inputs to minimize a number of comparison queries to be asked to the user, a comparison query requesting which outcome vector the user prefers between two outcome vectors representing possible outcomes of the healthcare decision problem, an outcome vector represented by multiple attributes associated with a treatment option;

elicit from a decision maker, based on the minimized number of comparison queries, a minimal amount of preference information with respect to the multiple attributes, wherein to elicit a minimal amount of preference information, said processor is further configured to;

identify N pairs of outcome vectors to present to the decision maker for preference assessment, said N pairs of outcome vectors identified by;

storing, in the memory device, a first set of input outcome vectors of the decision problem;

allocating, in said memory device, a memory structure defining a cone for storing a second set of N pairs of outcome vectors presented to the decision maker, said pairs of outcome vectors enumerated as a set (u, v);

estimating a decision maker preference of an outcome vector for a pair of outcome vectors, wherein if estimated that the decision maker prefers an outcome vector u over an outcome vector v, setting a score of a pair of outcome vectors (u, v) to be a number of undominated strategies obtained by solving the decision problem under an assumption that the outcome vector u is preferred to the outcome vector v;

otherwise, if estimated that the decision maker prefers an outcome vector v over an outcome vector u, setting a score of a pair of outcome vectors (v, u) to be a number of undominated strategies obtained by solving the decision problem under an assumption that the outcome vector v is preferred to the outcome vector u;

defining candidate pairs of outcome vectors as an ascending order of all the pairs of outcome vectors according to their corresponding scores;

for each candidate pair of the ordered outcome vector (u, v), confirming whether the decision maker prefers the outcome vector u over the outcome vector v, and in response to said confirming, adding the outcome vector pair (u, v) to the cone;

solving, using the processor, the decision problem with the current outcome vectors (u, v) stored in the cone;

evaluating whether the solution satisfies the decision maker;

if the decision maker does not satisfy the solution, generating outcome vectors u and v on a boundary of the cone by generating a linear combination of vectors already currently in the cone;

if the decision maker confirms that the outcome vector u is preferred over the outcome vector v, adding the pair of outcome vector (u, v) to the cone, otherwise, if an outcome vector v is preferred over the outcome vector u, adding the pair of outcome vector (v, u) to the cone; and

repeating the solving of the decision problem with the current cone, the evaluating, the generating, and the adding the pair of outcome vector (u, v) or outcome vector (v, u) to the cone until the solution satisfies the decision maker;

run the decision model to solve, based on the elicited minimal amount of preference information with respect to the multiple attributes, the decision problem within the pre-determined time period;

output, as the solution of the decision problem, a set of recommended actions;

wherein said decision maker makes a decision based on the set of recommended actions that avoids a full elicitation of preferences over all the multiple attributes.

7. The system according to claim 6, wherein the set of recommended actions are manageable by the decision maker so that the set of recommended actions satisfies a rule or is chosen by the decision maker.

8. The system according to claim 6 wherein prior to eliciting, implementing the decision model for determining a solution to a decision problem based on said multiple attributes and uncertainties of the decision problem, the decision problem including information about a plurality of outcome vectors that represent all possible outcomes and the uncertainties associated with the decision problem, the decision model precluding any preference information of the attributes and the outcome vectors.

9. The system according to claim 8, wherein the processor is further configured to:

determine whether the decision model can be solved without receiving any preference information from the decision maker.

10. The system according to claim 6, wherein the processor is further configured to:

present N elicitation questions, based on the identified N pairs of outcome vectors, to the decision maker, an elicitation question requesting of the decision maker which outcome vector the decision maker prefers between two outcome vectors; and

provide, by the decision maker, answers to the N elicitation questions as representing the minimal amount of preference information.

11. A computer program product for supporting a multi-attribute healthcare decision problem requiring a decision among possible treatment options, the computer program product comprising a non-transitory storage medium readable by a processing circuit and storing instructions run by the processing circuit for performing a method, said method steps comprising:

receiving, from a user, via an input to a computer system, a user'"'"'s indication of a maximum number of decision strategies from which the user is able to choose a strategy that solves the decision problem;

receiving, from a user, via the input to the computer system, a specification of an upper bound pre-determined time period within which a probabilistic decision tree model is run by a processor to solve the decision problem;

the processing circuit for running the probabilistic decision tree model based on the received inputs to minimize a number of comparison queries to be asked to a user, a comparison query requesting which outcome vector the user prefers between two outcome vectors representing possible outcomes of the decision problem, an outcome vector represented by multiple attributes associated with a treatment option;

eliciting from a decision maker, based on the minimized number of comparison queries, a minimal amount of preference information with respect to the multiple attributes, wherein said eliciting a minimal amount of preference information comprises;

identifying N pairs of outcome vectors to present to the decision maker for preference assessment, said N pairs of outcome vectors identified by;

storing, in a memory unit associated with the processing circuit, a first set of input outcome vectors of the decision problem;

allocating, in said memory unit, a memory structure defining a cone for storing a second set of N pairs of outcome vectors presented to the decision maker, said pairs of outcome vectors enumerated as a set (u, v);

estimating a decision maker preference of an outcome vector for a pair of outcome vectors, wherein if estimated that the decision maker prefers an outcome vector u over an outcome vector v, setting a score of a pair of outcome vectors (u, v) to be a number of undominated strategies obtained by solving the decision problem under an assumption that the outcome vector u is preferred to the outcome vector v;

otherwise, if estimated that the decision maker prefers an outcome vector v over an outcome vector u, setting a score of a pair of outcome vectors (v, u) to be a number of undominated strategies obtained by solving the decision problem under an assumption that the outcome vector v is preferred to the outcome vector u;

defining candidate pairs of outcome vectors as an ascending order of all the pairs of outcome vectors according to their corresponding scores;

for each candidate pair of the ordered outcome vector (u, v), confirming whether the decision maker prefers the outcome vector u over the outcome vector v, and in response to said confirming, adding the outcome vector pair (u, v) to the cone;

solving, using the processing circuit, the decision problem with the current outcome vectors (u, v) stored in the cone;

evaluating whether the solution satisfies the decision maker;

if the decision maker does not satisfy the solution, generating outcome vectors u and v on a boundary of the cone by generating a linear combination of vectors already currently in the cone;

if the decision maker confirms that the outcome vector u is preferred over the outcome vector v, adding the pair of outcome vector (u, v) to the cone, otherwise, if an outcome vector v is preferred over the outcome vector u, adding the pair of outcome vector (v, u) to the cone; and

repeating the solving of the decision problem with the current cone, the evaluating, the generating, and the adding the pair of outcome vector (u, v) or outcome vector (v, u) to the cone until the solution satisfies the decision maker;

solving by running the decision model, based on the elicited minimal amount of preference information with respect to the multiple attributes, the decision problem within the pre-determined time period;

outputting, as the solution of the decision problem, a set of recommended actions;

wherein said decision maker makes a decision based on the set of recommended actions that avoids a full elicitation of preferences over all the multiple attributes.

12. The computer program product according to claim 11, wherein the set of recommended actions are manageable by the decision maker so that the set of recommended actions satisfies a rule or is chosen by the decision maker.

13. The computer program product according to claim 11 wherein prior to eliciting, implementing the decision model for determining a solution to a decision problem based on said multiple attributes and uncertainties of the decision problem, the decision problem including information about a plurality of outcome vectors that represent all possible outcomes and the uncertainties associated with the decision problem, the decision model precluding any preference information of the attributes and the outcome vectors.

14. The computer program product according to claim 13, wherein the instructions configure the processing circuit to:

determine whether the decision model can be solved without receiving any preference information from the decision maker.

15. The computer program product according to claim 11, wherein the instructions further configure the processing circuit to:

present N elicitation questions, based on the identified N pairs of outcome vectors, to the decision maker, an elicitation question requesting of the decision maker which outcome vector the decision maker prefers between two outcome vectors; and

provide, by the decision maker, answers to the N elicitation questions as representing the partially specified preference information.