Hierarchical Rule Clustering for Machine Learning Model Understanding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques make it difficult for users to easily grasp the contents of complex machine learning models, particularly those with numerous rules, leading to increased workload and time in understanding their operation.
Innovation Solution
An information processing device that calculates feature values for rules, classifies them into clusters based on similarity, identifies inclusion and hierarchical relationships between clusters, and outputs graphical information to intuitively represent these relationships, thereby simplifying the understanding of the model's structure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning models with multiple rules are generated to classify data, then classification accuracy and model capability are improved, but user understanding and model interpretability deteriorate
Solution Approach 1:
The patent segments the complex rule set into multiple clusters based on similarity of feature values. Each cluster groups rules with comparable characteristics, making the overall model structure more manageable and interpretable while preserving the complete classification capability of all rules.
Solution Approach 2:
The patent introduces a hierarchical dimension by organizing clusters into parent-child relationships. This transforms the flat, complex rule set into a multi-level structure where users can understand the model at different granularities - from high-level cluster relationships to detailed individual rules - thereby improving interpretability without losing accuracy.
2Adaptability or versatility
If the number of rules in the machine learning model increases to handle complex classification tasks, then model functionality is improved, but the time and workload required to understand the model increases
Solution Approach 1:
By dividing the large rule set into smaller clusters, the patent reduces the cognitive load on users. Instead of examining every individual rule, users can first understand the cluster-level patterns and then drill down into specific clusters only when needed, significantly reducing the time required to comprehend the overall model functionality.
Solution Approach 2:
The patent performs preliminary organization of rules into clusters and identification of hierarchical relationships before user inspection. This pre-processing creates an optimized view of the model structure that allows users to quickly grasp the essential classification logic without being overwhelmed by the full complexity of all rules.
3Ease of operation
If detailed rule conditions are provided to enhance model transparency, then interpretability is improved, but the complexity of the information presented increases
Solution Approach 1:
The patent applies local quality by providing different levels of detail in different parts of the model representation. At the cluster level, summary information is provided to avoid overwhelming complexity. At the individual rule level within clusters, detailed conditions are available for those who need to understand specific classification logic, thus balancing transparency with manageable complexity.
Solution Approach 2:
By organizing rules into hierarchical clusters, the patent creates a multi-dimensional view where information complexity is managed through levels of abstraction. Users can navigate from simplified cluster-level descriptions to detailed rule-level information as needed, transforming a single-complexity-level problem into a multi-level structure that reduces perceived complexity.
Data Source
AI summary
An information processing device generates a feature vector for each rule represented by a model. The information processing device classifies each rule represented by the model into any one of multiple clusters based on the Euclidean distance between the generated feature vectors. The information processing device identifies an inclusion relationship between clusters among the multiple clusters. The information processing device identifies a hierarchical relationship between clusters based on the inclusion relationships between clusters. The information processing device displays a graph representing the identified hierarchical relationship between clusters. In response to a designation of any node in a displayed graph, the information processing device displays explanatory information related to one or more rules classified into a cluster represented by the designated any node.