Reproduce “ComRAG: Retrieval-Augmented Generation with Dynamic Vector Stores for Real-time Community Question Answering in Industry” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Qinwen Chen, Wenbiao Tao, Zhiwei Zhu, Mingfan Xi, Liangzhong Guo, Yuan Wang, Wei Wang, Yunshi Lan
ComRAG is a Retrieval-Augmented Generation framework that integrates static domain knowledge with dynamic, quality-aware historical Q&A data for industrial Community Question Answering. It achieved superior answer quality and reduced query latency by up to 54% in real-time scenarios while efficiently managing memory.
Key insights
- Integrating static domain expertise with dynamic, quality-classified historical community interactions significantly enhances both the accuracy and real-time efficiency of CQA systems.
- A centroid-based memory mechanism allows for scalable and efficient management of continuously growing historical Q&A data by maintaining representative clusters and prioritizing high-quality information.
- Leveraging low-quality historical answers as 'negative examples' during LLM generation can guide the model to avoid common pitfalls and generate more robust responses.
- Dynamically adjusting LLM generation temperature based on the consistency of retrieved evidence improves the appropriateness of generated answers, balancing exploration and determinism.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.