Semantic Textual Similarity Search with Entity-Aware Explainability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semantic textual similarity search methods rely solely on token or sentence encoding, failing to provide transparent and effective similarity measurements that consider important entities and their context, making it difficult for human users to understand the reasoning behind similarity outcomes.
Innovation Solution
A computer-implemented method that identifies and classifies name entities, generates semantic textual similarity models, aggregates entity embeddings with structural features, and uses a similarity estimator to find similar documents while providing explanatory information on the degree of similarity, leveraging both textual and entity-based information for enhanced performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If token or sentence encoding is used for similarity measurement, then the overall textual semantic meaning is captured, but important entities and their context are not adequately considered
Solution Approach 1:
The patent segments the text processing into multiple stages: initial token/sentence encoding, entity identification and extraction, entity embedding generation, and final aggregation. This segmentation allows each component to focus on specific aspects (general semantics vs. entity-specific information) without losing important details, thereby resolving the contradiction between capturing overall meaning and preserving entity information.
Solution Approach 2:
The patent merges multiple encoding approaches by combining the initial token/sentence embeddings with entity-specific embeddings. The final document representation integrates both general textual semantics and entity-level details, achieving comprehensive similarity measurement that captures both overall meaning and important entity context simultaneously.
2Productivity
If AI systems find similar documents based on semantic encoding, then similarity outcomes are generated, but human users cannot understand how and why the outcome was generated
Solution Approach 1:
The patent introduces an explainability generator as an intermediary component that bridges the AI system's internal reasoning process and human understanding. This generator creates human-interpretable explanations by analyzing the entity-level similarities and differences, translating complex embedding comparisons into understandable rationale without affecting the efficiency of the core similarity search process.
Solution Approach 2:
The system segments the similarity analysis into traceable components at the entity level, allowing the explainability generator to provide detailed breakdowns of which specific entities contributed to the similarity score. This segmentation makes the AI's reasoning process transparent and understandable to human users while maintaining search efficiency.
3Measurement precision
If entity-based encoding is added to improve similarity measurement, then entity context is better captured, but system complexity increases
Solution Approach 1:
The patent applies preliminary action by performing entity identification and extraction before the main similarity computation. Entities are identified, classified, and embedded in advance, allowing the core similarity search to work with pre-processed entity representations rather than raw text. This preliminary processing organizes the complexity into manageable stages without significantly increasing the overall system burden.
Solution Approach 2:
The patent employs universal components that serve multiple functions: the entity extractor not only identifies entities but also classifies them; the embedding stage generates both entity embeddings and contributes to document-level embeddings; the aggregation module combines entity and structural features while maintaining compatibility with standard similarity search operations. This multi-functionality reduces the need for separate specialized components, managing system complexity.
Data Source
AI summary
A method of performing a semantic textual similarity search between a target document and a set of source documents includes: selecting target textual data from the target document; identifying and classifying, name entities for each textual sequence of the target textual data; generating, a semantic textual similarity model and using the semantic textual similarity model to generate textual embedding of the identified name entities for each textual sequence of the target textual data; aggregating the textual embedding with the structural features of the target textual data to generate target document embedding; searching, by a similarity estimator, for similar documents by measuring similarities of the target document embedding with embeddings of each source document of the set of source documents; and computing explanatory information about a degree of similarity between the target document and any document of the set of the source documents.