LSA Vector Propagation for Concept Similarity in QA Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for similarity estimation between concepts in knowledge bases are limited, as they either do not account for distributional properties of concepts or lack sufficient coverage, and struggle with word sense disambiguation, which hampers effective answer scoring in question-answering systems.
Innovation Solution
A vector propagation algorithm is employed to improve similarity estimation by propagating Latent Semantic Analysis (LSA) vectors through a graph structure derived from a knowledge base, incorporating the strength of relations between concepts to generate modified similarity measures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If graph algorithms are used to estimate similarity between concepts, then the method can handle concepts not in the corpus, but it does not take into account distributional properties of concepts such as frequency
Solution Approach 1:
The patent combines graph algorithms with corpus-based algorithms by propagating LSA vectors through the knowledge base graph. This merging allows the system to leverage both the coverage advantage of graph-based approaches and the precision advantage of corpus-based distributional analysis, resolving the contradiction between coverage and measurement precision.
Solution Approach 2:
The patent introduces LSA vectors as an intermediary representation that bridges the knowledge base graph and corpus data. These vectors serve as a mediator that captures distributional properties from the corpus while being propagated through the graph structure, enabling simultaneous achievement of coverage and precision.
2Measurement precision
If corpus based algorithms are used to estimate similarity between concepts, then distributional properties are considered, but coverage is insufficient as not all terms in the ontology are in the corpus
Solution Approach 1:
The patent performs preliminary computation of LSA vectors for concepts in the corpus, then uses these pre-computed vectors to propagate through the knowledge base graph. This preliminary action enables the system to leverage corpus-based precision for concepts that are actually in the corpus while extending coverage to concepts not in the corpus through graph propagation.
Solution Approach 2:
The patent introduces LSA vectors as an intermediary that connects corpus-based precision with graph-based coverage. The vectors serve as a bridge that allows precise similarity estimation for corpus terms while enabling inference for terms not in the corpus through graph propagation.
3Productivity
If standard LSA is applied to estimate similarity, then computational efficiency is maintained, but it cannot effectively handle word sense disambiguation in specific context domains
Solution Approach 1:
The patent applies local quality by propagating LSA vectors through the knowledge base graph with respect to specific context domains. The propagation process respects the local structure and semantics of the graph, allowing different concepts to be differentiated based on their specific contextual relationships, thereby improving word sense disambiguation while maintaining computational efficiency.
Data Source
AI summary
A system and method that improves obtaining similarity measure between concepts based on Latent Semantic Analysis by taking onto account graph structure derived from the knowledge bases by using a vector propagation algorithm, in the context domain, such as a medical domain. Concepts contained in a corpus of documents are expressed in a graph wherein each node is a concept and edges between node express relation between concepts weighted by the number of semantic relations determined from the corpus. A vector of neighbors is created and assigned to each concept, thereby providing an improved similarity measure between documents, i.e., corpus and query against corpus.