HeteRAG: A Heterogeneous Retrieval-augmented Generation Framework with Decoupled Knowledge Representations
This addresses a specific bottleneck in RAG systems for improving LLM performance, though it appears incremental as it builds on existing RAG methods.
The paper tackles the problem that retrieval-augmented generation (RAG) methods use identical knowledge representations for both retrieval and generation, which is suboptimal because retrieval benefits from comprehensive information while generation suffers from redundant context. The proposed HeteRAG framework decouples these representations, achieving significant improvements in effectiveness and efficiency over baselines.
Retrieval-augmented generation (RAG) methods can enhance the performance of LLMs by incorporating retrieved knowledge chunks into the generation process. In general, the retrieval and generation steps usually have different requirements for these knowledge chunks. The retrieval step benefits from comprehensive information to improve retrieval accuracy, whereas excessively long chunks may introduce redundant contextual information, thereby diminishing both the effectiveness and efficiency of the generation process. However, existing RAG methods typically employ identical representations of knowledge chunks for both retrieval and generation, resulting in suboptimal performance. In this paper, we propose a heterogeneous RAG framework (\myname) that decouples the representations of knowledge chunks for retrieval and generation, thereby enhancing the LLMs in both effectiveness and efficiency. Specifically, we utilize short chunks to represent knowledge to adapt the generation step and utilize the corresponding chunk with its contextual information from multi-granular views to enhance retrieval accuracy. We further introduce an adaptive prompt tuning method for the retrieval model to adapt the heterogeneous retrieval augmented generation process. Extensive experiments demonstrate that \myname achieves significant improvements compared to baselines.