Query Term Expansion Using Weighted Mutual Affinity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing query result retrieval techniques face challenges in efficiently reducing irrelevant results due to the inclusion of homonyms and unrelated expansion terms, which are not semantically relevant to the query subject, often requiring significant computational resources and data storage.
Innovation Solution
The implementation of a Weighted Mutual Affinity (WMA) method to determine the semantic similarity between query and expansion terms, using word embeddings and belief networks to cull irrelevant terms and rank documents based on their relevance, thereby improving the accuracy of query result retrieval.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a thesaurus is used to expand query terms, then query result retrieval is improved, but significant data storage is required
Solution Approach 1:
The patent extracts and utilizes only the necessary semantic relationships from the thesaurus structure, specifically focusing on hypernym relationships and term co-occurrence patterns rather than storing the complete thesaurus. This allows query expansion functionality to be maintained while significantly reducing the data storage requirements by extracting only the essential semantic information needed for accurate query result retrieval.
2Reliability
If graph clustering is used to expand query terms, then query result retrieval is improved, but more compute bandwidth is required
Solution Approach 1:
The patent segments the query expansion process into distinct phases: first identifying candidate expansion terms based on pre-computed semantic relationships, then performing targeted clustering only on relevant term groups. This segmentation reduces the overall compute bandwidth requirement by avoiding full-graph clustering operations and focusing computational resources only on the specific subset of terms relevant to the query.
Solution Approach 2:
The patent performs preliminary computation of semantic relationships and term co-occurrence statistics during an indexing phase, storing these pre-computed values for quick retrieval during query processing. This preliminary action eliminates the need for expensive real-time graph clustering operations, significantly reducing compute bandwidth requirements during actual query result retrieval while maintaining high accuracy.
3Reliability
If augmented word embedding model is used to compute result relevance, then query result retrieval is improved, but computational resources are increased
Solution Approach 1:
The patent changes the parameters of word embedding by using pre-computed, fixed-dimensional embeddings rather than dynamically generating them through complex augmented models. By pre-calculating and storing embedding vectors with optimized dimensions, the system maintains high relevance computation accuracy while significantly reducing the computational resources required during query processing, as the embeddings can be quickly retrieved and compared without running intensive training or generation operations.
4Reliability
If deep learning with attention is used, then query result retrieval is improved, but computational complexity is increased
Solution Approach 1:
The patent replaces complex deep learning models with simpler, pre-computed semantic representations that can be quickly applied during query processing. Instead of using computationally intensive deep learning with attention mechanisms, the system employs pre-generated semantic vectors and relationship graphs that provide sufficient accuracy with much lower computational complexity, effectively using simpler 'disposable' computational structures rather than maintaining complex persistent models.
Data Source
AI summary
Devices, systems, and methods for improving results returned from a query. A method can include identify, based on a term embedding of a corpus of terms, expansion terms of a raw query term that are nearest the raw query term, normalize distances between the raw query term and the identified expansion terms, identify, based on the term embedding, expansion term neighbors of an expansion term that are nearest the expansion term; normalize distances between the expansion term and the identified expansion term neighbors, determine a WMA weight between the raw query term and the expansion term, and execute the query with the raw query terms and the expansion terms (determined based on the WMA weight) to generate query results.