Cross-Document Coreference via Hierarchical Vector Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current natural language processing technologies face challenges in accurately performing cross-document coreference, particularly in identifying and clustering mentions that refer to the same entity across multiple documents, due to ambiguities and the reliance on pre-enumerated aliases or variants, which limits their effectiveness in applications like clinical decision support and web search.
Innovation Solution
A computer-implemented method using machine learning algorithms to determine mentions in input documents by parsing them into spelling and context vectors, generating a hierarchical tree data structure, and iteratively merging nodes with high similarity scores to create entity nodes, thereby identifying coreferent mentions and constructing robust entity representations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional NLP techniques are used for cross-document coreference, then the system can identify mentions using pre-enumerated aliases, but the accuracy deteriorates due to ambiguities and limited effectiveness in complex applications
Solution Approach 1:
The patent transforms discrete mention identification into continuous vector space representation. By mapping mentions to dense vector embeddings that capture semantic meaning rather than relying on predefined aliases, the system achieves both higher accuracy and better adaptability to complex applications like clinical decision support and web search.
Solution Approach 2:
The patent replaces rule-based alias matching with machine learning-based vector similarity computation. This substitution enables the system to handle ambiguities and edge cases that plague traditional approaches, improving performance in complex real-world applications.
2Measurement precision
If mentions are represented using only spelling data, then the processing is simple, but the coreference resolution accuracy deteriorates due to lack of contextual information
Solution Approach 1:
The patent merges spelling-based representations with context-based representations into a unified vector space. By combining lexical information from spellings with semantic information from contexts, the system achieves accurate coreference resolution while maintaining manageable complexity through integrated representation.
Solution Approach 2:
The patent creates composite mention representations by combining multiple data sources (spelling vectors and context vectors). This composite approach leverages the strengths of both lexical matching and contextual understanding to improve coreference accuracy without overwhelming complexity.
3Measurement precision
If all mentions are treated equally in the corpus, then the processing is uniform, but the entity clustering quality deteriorates due to noise and salient mention identification
Solution Approach 1:
The patent applies different weighting to different mentions based on their local characteristics and importance. By identifying salient mentions and assigning them higher weights while downweighting noisy mentions, the system improves entity clustering quality without requiring complex global processing.
Solution Approach 2:
The patent segments the mention representation into distinct components (spelling and context vectors) that can be processed and weighted independently. This segmentation allows selective attention to salient features while filtering noise, improving clustering quality with manageable complexity.
Data Source
AI summary
A computer-implemented method for performing cross-document coreference for a corpus of input documents includes determining mentions by parsing the input documents. Each mention includes a first vector for spelling data and a second vector for context data. A hierarchical tree data structure is created by generating several leaf nodes corresponding to respective mentions. Further, for each node, a similarity score is computed based on the first and second vectors of each node. The hierarchical tree is populated iteratively until a root node is created. Each iteration includes merging two nodes that have the highest similarity scores and creating an entity node instead at a hierarchical level that is above the two nodes being merged. Further, each iteration includes computing the similarity score for the entity node. The nodes with the similarity scores above a predetermined value are entities for which coreference has been performed in input documents.