Dynamic Tokenization for Entity Matching Explanations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing explainable AI (XAI) technologies struggle to provide transparent and understandable explanations for non-technical users, particularly in critical applications like medical diagnosis and investment decisions, as they often output ambiguous results that fail to instill trust in the reliability of machine learning model predictions.
Innovation Solution
A machine learning system that uses dynamic sub-word-level tokenization to generate explanations for entity matching, providing a reason for the match by determining character-level embeddings, sub-word-level tokens, and attention scores, which are then used to create an attention matrix to output clear explanations for query and target entities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional XAI methods are used to provide explanations for ML model predictions, then some level of transparency is achieved, but the explanations remain ambiguous and difficult for non-technical users to understand
Solution Approach 1:
The patent segments the explanation generation process into distinct components: (1) obtaining the prediction output from the ML model, (2) generating multiple candidate explanations through different reasoning paths, (3) evaluating each candidate explanation's quality and relevance, and (4) selecting the best explanation to present to the user. This segmentation allows each component to be optimized independently, improving both reliability and understandability.
Solution Approach 2:
The patent introduces an intermediary explanation generation system that acts as a mediator between the black-box ML model and the end user. This intermediary component translates the model's internal decision-making process into multiple candidate natural language explanations, which are then evaluated and selected to provide clear, understandable justifications that bridge the gap between technical model operations and non-technical user comprehension.
2Reliability
If detailed explanations are provided to improve transparency, then trust may be enhanced, but the complexity of the explanation system increases
Solution Approach 1:
The patent generates multiple candidate explanations (excessive action) but then selects only the most relevant and clear ones for presentation (partial action). This approach ensures comprehensive coverage of possible reasoning paths while avoiding information overload, maintaining transparency without requiring the system to present all possible explanations, thus managing complexity effectively.
Solution Approach 2:
The patent implements a dynamic explanation selection process that adapts to the specific prediction context. The system evaluates candidate explanations based on their relevance, clarity, and appropriateness for the given prediction, dynamically selecting the best explanation rather than using a fixed template. This dynamic approach enhances transparency while keeping the system manageable through context-aware selection.
Data Source
AI summary
Methods, systems, and computer-readable storage media for receiving query data representative of query entities and target data representative of target entities, determining, by an attention ML model, a set of character-level embeddings, providing, by a sub-word-level tokenizer, a set of sub-word-level tokens, each sub-word-level token including a string of multiple characters, generating, by the attention ML model, a set of sub-word-level embeddings based on the set of sub-word-level tokens, providing, by the attention ML model, at least one attention matrix including attention scores, each attention score representative of a relative importance of a respective sub-word-level token in a predicted match, the predicted match including a match between a query entity and a target entity, and outputting an explanation based on the at least one attention matrix.