RAG Query Expansion via Vector Search and Reciprocal Rank Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Retrieval Augmented Generation (RAG) systems face challenges in effectively retrieving information due to limitations in current search technologies, which struggle with capturing the nuances and contexts of complex user queries, leading to less relevant results.
Innovation Solution
The system generates a plurality of natural language queries based on an original query, performs vector searches for these queries, and uses reciprocal rank fusion and custom vector score weighting to re-rank results, thereby providing comprehensive and accurate search outcomes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If current search technologies are used for information retrieval, then the system is simple and easy to operate, but the search results are less relevant and fail to capture the nuances and contexts of complex user queries
Solution Approach 1:
The patent segments the query processing into multiple components: query understanding module, query expansion module, vector search module, and result re-ranking module. Each component handles a specific aspect of the search process, allowing the system to capture nuanced meanings and contexts while maintaining modular architecture that manages complexity.
Solution Approach 2:
The patent introduces an intermediary vector embedding layer that transforms queries and documents into vector representations. This intermediary representation enables the system to capture semantic similarities and contextual relationships between queries and documents, improving search relevance without requiring direct complex analysis of the original query text.
2Measurement precision
If multiple queries are generated and vector searches are performed, then the search result quality and depth are enhanced, but the processing time and computational resources increase
Solution Approach 1:
The patent performs preliminary actions by pre-computing vector embeddings for documents and maintaining an indexed structure. When a query arrives, the system can quickly retrieve relevant documents using pre-computed vectors without needing to perform complex analysis in real-time, thus reducing processing time while maintaining high search quality.
Solution Approach 2:
The patent replaces traditional mechanical search processes with vector-based semantic search. Instead of relying on keyword matching and traditional indexing mechanisms, the system uses vector embeddings to capture semantic relationships, enabling faster and more accurate retrieval of relevant information through computational geometry operations.
3Loss of information
If traditional search technologies are used, then the system is efficient and fast, but it fails to capture the intended meaning behind user queries and provides superficial results
Solution Approach 1:
The patent changes the fundamental parameter of query representation from discrete keywords to continuous vector embeddings. This parameter change enables the system to capture nuanced meanings, synonyms, and contextual relationships that traditional keyword-based approaches miss, while maintaining search efficiency through optimized vector search algorithms.
Data Source
AI summary
Systems and methods herein provide a processor; and a non-transitory, processor readable storage medium communicatively coupled to the processor. The non-transitory, processor readable storage medium may include one or more instructions stored thereon that, when executed, cause the processor to: input one or more queries into a large language model; generate, based on the one or more queries, a plurality of natural language queries, wherein each of the plurality of natural language queries are distinct queries and associated with the one or more queries; perform vector searches for the one or more queries and plurality of natural language queries; compile the plurality of natural language queries into a search result based on the vector searches; and generate a summary based on the search result.