Question Style Rewriting for More Accurate RAG Retrieval
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional retrieval-augmented generation systems face challenges in maintaining retrieval accuracy and naturalness of answers due to differences in question sentence expressions and technical terms, leading to decreased accuracy in generating answers.
Innovation Solution
A method that involves modifying the style of a question sentence using a language model to match the style of stored sentence data, thereby improving retrieval accuracy by generating a second question sentence that is more likely to hit appropriate information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a conventional retrieval-augmented generation system uses a question sentence as input to retrieve information from a database, then the system can generate answers based on retrieved information, but the retrieval accuracy decreases when there are differences in question sentence expressions and technical terms
Solution Approach 1:
The system changes the parameters of the question sentence by modifying its expression style and technical terms to match those in the database. A language model generates alternative versions of the question sentence with different expressions while maintaining the original meaning, thereby improving retrieval accuracy when the database contains information in specific expression patterns
Solution Approach 2:
The system creates multiple copies of the question sentence with different expressions generated by a language model. These copied versions are used to retrieve information from the database, increasing the likelihood of finding matching information even when the original question expression differs from database entries
2Measurement precision
If the system modifies the style of a question sentence to match stored data style, then retrieval accuracy improves, but the complexity of the system increases due to additional processing steps
Solution Approach 1:
The system introduces a language model as an intermediary component that transforms the original question sentence into multiple versioned expressions. This mediator handles the complexity of style matching and expression generation, allowing the core retrieval system to focus on searching with improved query variants
Solution Approach 2:
The system performs preliminary action by generating multiple expression versions of the question sentence before the actual retrieval process. This pre-processing step creates optimized query variants that are more likely to match database entries, thereby improving retrieval accuracy without complicating the core search mechanism
Data Source
AI summary
A computer-readable recording medium stores therein a generation program for causing a computer to execute a process including: retrieving a first sentence data related to a first question sentence by referring to a storage unit that stores a plurality of sentence data; and generating a second question sentence in which a style of the first question sentence is modified so as to maintain a meaning of the first question sentence, the second question sentence being generated based on the retrieved first sentence data, using a first language model for generating a sentence.