Query Expansion Graph Using Spreading Activation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In information retrieval systems, query expansion techniques often fail to effectively retrieve relevant documents, especially in domain-specific searches like legal, financial, or biomedical, where the user's query text may not match the expected answer text, and traditional methods rely on semantic relationships that are not always apparent.
Innovation Solution
A computer-implemented method and system that generates a graph of question and answer vocabulary using disjoint word pairs from training data, applying spreading activation to identify candidate words for query expansion, and updates the graph based on user feedback to refine search results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional query expansion techniques are used to find synonyms and related terms, then the search query coverage is improved, but the retrieval precision deteriorates in domain-specific searches where user query text does not match expected answer text
Solution Approach 1:
The patent segments the query expansion process into two distinct phases: (1) building a domain-specific vocabulary graph from training question-answer pairs that captures non-obvious semantic relationships, and (2) applying spreading activation on this pre-built graph to expand user queries. This segmentation allows the system to maintain a rich domain vocabulary structure while applying controlled, context-aware expansion only when needed, thus improving both coverage and precision.
Solution Approach 2:
The patent performs preliminary action by pre-building the vocabulary graph from training data before actual query processing. The graph is constructed in advance with domain-specific terms and their relationships, so that when a user query arrives, the system can quickly apply spreading activation without performing heavy semantic analysis during query time. This preliminary preparation ensures both comprehensive domain coverage and precise, controlled expansion.
2Reliability
If spreading activation is applied through the graph to expand query terms, then the retrieval of relevant documents is improved, but the complexity of the query processing system increases
Solution Approach 1:
The vocabulary graph is built in advance from training question-answer pairs, pre-computing all semantic relationships between domain terms. This preliminary construction phase captures complex domain knowledge that would be difficult to compute during query processing. During actual query expansion, only the relatively simple spreading activation algorithm needs to be applied, which propagates activation values through pre-established edges in the graph.
Solution Approach 2:
The vocabulary graph serves as an intermediary structure between user queries and the document collection. Instead of directly analyzing semantic relationships between queries and documents (which would be computationally intensive), the system uses the pre-built graph as a mediator. The graph captures domain-specific semantic relationships and uses spreading activation to efficiently bridge the gap between user terminology and document content.
3Measurement precision
If disjoint answer word nodes are used in the graph to capture non-obvious relationships, then the semantic relationship detection is improved, but the graph structure complexity increases
Solution Approach 1:
The patent extracts only the necessary elements for capturing non-obvious semantic relationships: disjoint answer word nodes (words that appear in answers but not in corresponding questions) and the edges connecting them to question words. By taking out only these critical components rather than including all possible word relationships, the graph maintains sufficient complexity to detect non-obvious semantic relationships while avoiding unnecessary structural complexity.
Solution Approach 2:
The graph applies local quality by focusing computational and structural resources on capturing non-obvious relationships specifically at the disjoint answer word nodes. These nodes receive special treatment in the graph structure, with edges connecting them to question words that capture the non-obvious semantic mappings. This localized approach ensures high precision in detecting non-obvious relationships without requiring the entire graph to be overly complex.
Data Source
AI summary
A method and/or system for query expansion may include: providing a set of training data in a given domain in the form of training question texts and training answer texts, identifying disjoint answer words in the training answer text that do not occur in the associated training question text, generating a graph of question word nodes and answer word nodes generated from the set of training data for the given domain in the form of the training question texts and the training answer texts, wherein edges are provided between a disjoint pair of a question word node for a question word in a training question and an answer word node for a disjoint answer word in an associated training answer, and applying spreading activation through the graph to result in a top n most highly activated nodes that are used as candidate words for expansion of a user query input.