Score Explanation Method Using Shapley Values
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for explaining scores calculated by decision-making systems are often computationally intensive, less accurate, and not easily adaptable to different mathematical models, making it difficult for decision-makers to understand which parameters significantly influence the scoring process.
Innovation Solution
A method that generates explanations for scores associated with vectors by using a function, which includes providing a score, generating lists of indicators, calculating alternative vectors, evaluating significance indicators such as Shapley values or Sobol' indices, and elaborating an explanation from these indicators, allowing for adaptability to various mathematical models and improved relevance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing explanation methods are applied to generate score explanations, then the explanation can be obtained, but the computing time required is too high
Solution Approach 1:
The patent segments the explanation generation process into multiple independent steps: generating alternative vectors by modifying individual parameters, calculating scores for each alternative vector, computing significance indicators (Shapley values or Sobol' indices) for each parameter, and synthesizing the final explanation. This segmentation allows for optimized computation at each stage and enables parallel processing of different parameter variations, significantly reducing overall computing time while maintaining explanation accuracy.
Solution Approach 2:
The patent implements a configurable approach where the system can generate a limited number of alternative vectors (e.g., top K most significant ones) rather than exhaustively evaluating all possible parameter combinations. This partial action approach provides sufficiently accurate explanations with dramatically reduced computational resources, allowing decision-makers to obtain actionable insights without waiting for complete exhaustive analysis.
2Adaptability or versatility
If the mathematical model complexity increases to handle more parameters, then the scoring capability improves, but the global score becomes less understandable
Solution Approach 1:
The patent introduces significance indicators (Shapley values or Sobol' indices) as intermediary metrics that bridge the complex mathematical model and the decision-maker's understanding. These indicators quantify each parameter's contribution to the final score in an intuitive manner, allowing decision-makers to understand which parameters drove the score up or down without needing to comprehend the underlying complex mathematical relationships. The system acts as a translator between model complexity and human interpretability.
Solution Approach 2:
The patent segments the complex score into attributable components by calculating individual significance indicators for each parameter. Instead of presenting a single opaque global score, the system breaks down the scoring result into parameter-specific contributions, making it clear which factors positively or negatively influenced the outcome. This segmentation transforms an intimidating complex model output into digestible, actionable information for decision-makers.
3Adaptability or versatility
If existing explanation methods are used, then the explanation can be generated, but it is not easily adaptable to systems using other mathematical models
Solution Approach 1:
The patent employs universal significance indicator calculations (Shapley values or Sobol' indices) that can be applied to any mathematical model producing a score from multiple parameters. Rather than developing model-specific explanation methods, the system uses a unified approach that works with linear models, non-linear models, machine learning models, or any other scoring function. This universal methodology significantly enhances adaptability across different domains and model types while maintaining relatively simple implementation through standardized calculation procedures.
Data Source
AI summary
A score explanation method for explaining a score includes at least steps of: a1) providing a first score associated with a first vector containing the first values of the parameters; b) generating a first set of lists, each list including a second number of indicators; c) generating, from a list, of at least a third vector wherein each parameter has a third value; the third value being equal to the corresponding first value when the list does not include an indicator of the corresponding parameter, and different from the corresponding first value otherwise; d) calculating the score of at least one third vector; e) evaluating, from the scores calculated for each of the third vectors, an indicator of significance of each parameter; f) elaborating, from the evaluated indicators of significance, an explanation of the first score.