Learning policy explanations

A system for learning policy explanations in a computing environment, comprising:

• one or more processors with executable instructions that when executed cause the system to:
• ingest text data from a policy data source into a data repository;
• execute machine learning logic to generate a policy rule model trained according to one or more rules extracted from the text data, wherein a classifier of the policy rule model identifies one or more segments of the text having clause-based obligatory language notwithstanding whether the clause-based obligatory language explicitly describes a rule, classifies the clause-based obligatory language as one or more rules, and determines bounds of the one or more rules outside of which the one or more rules are violated;
• responsive to receiving input of a claim via a user interface, provide, as output on the user interface, one or more explanations justifying validity or invalidity of the claim based on the one or more rules extracted from the one or more segments of the text data of the policy data source according to the executed machine learning logic;
• collect feedback from a user for learning a degree of relevance of the one or more rules for inclusion or exclusion in the one or more explanations;
• assign a cost to each of the one or more rules associated with the claim, and optimizing the cost by performing a value update operation according to the collected feedback, wherein the cost is indicative as to an importance that the one or more rules are utilized for the one or more explanations, and those of the one or more rules having a smallest cost subsequent to the value update operation are chosen as being referenced as a best explanation of the one or more explanations to provide to the user, and wherein, upon the user selecting a first provided explanation of the one or more explanations, the value update operation decreases the cost of a first rule of the one or more rules associated with the first provided explanation and increases the cost of a second rule associated with a second, unselected explanation of the one or more explanations until the costs of each of the one or more rules are updated to a threshold level;
• and
• execute the machine learning logic to iteratively update the policy rule model using the feedback collected from the user, wherein iteratively updating the policy rule model increases an accuracy in identifying and generating the one or more rules and the one or more explanations.

A computer program product for, by one or more processors, learning policy explanations in a computing environment, the computer program product comprising a non-transitory computer-readable storage medium having computer-readable program code portions stored therein, the computer-readable program code portions comprising:

• an executable portion that ingests text data from a policy data source into a data repository;
• an executable portion that executes machine learning logic to generate a policy rule model trained according to one or more rules extracted from the text data, wherein a classifier of the policy rule model identifies one or more segments of the text having clause-based obligatory language notwithstanding whether the clause-based obligatory language explicitly describes a rule, classifies the clause-based obligatory language as one or more rules, and determines bounds of the one or more rules outside of which the one or more rules are violated;
• an executable portion that, responsive to receiving input of a claim via a user interface, providing, as output on the user interface, one or more explanations justifying validity or invalidity of the claim based on the one or more rules extracted from the one or more segments of the text data of the policy data source according to the executed machine learning logic;
• an executable portion that collects feedback from a user for learning a degree of relevance of the one or more rules for inclusion or exclusion in the one or more explanations;
• an executable portion that assigns a cost to each of the one or more rules associated with the claim, and optimizing the cost by performing a value update operation according to the collected feedback, wherein the cost is indicative as to an importance that the one or more rules are utilized for the one or more explanations, and those of the one or more rules having a smallest cost subsequent to the value update operation are chosen as being referenced as a best explanation of the one or more explanations to provide to the user, and wherein, upon the user selecting a first provided explanation of the one or more explanations, the value update operation decreases the cost of a first rule of the one or more rules associated with the first provided explanation and increases the cost of a second rule associated with a second, unselected explanation of the one or more explanations until the costs of each of the one or more rules are updated to a threshold level;
• and
• an executable portion that executes the machine learning logic to iteratively update the policy rule model using the feedback collected from the user, wherein iteratively updating the policy rule model increases an accuracy in identifying and generating the one or more rules and the one or more explanations.