Reproduce “Synergistic Interplay between Search and Large Language Models for Information Retrieval” 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
Jiazhan Feng, Chongyang Tao, Xiubo Geng, Tao Shen, Can Xu, Guodong Long, Dongyan Zhao, Daxin Jiang
The InteR framework introduces an iterative information refinement process that synergistically combines Retrieval Models and Large Language Models for information retrieval. This approach achieved state-of-the-art zero-shot retrieval performance, outperforming previous zero-shot baselines by over 8% absolute MAP gain and often surpassing supervised fine-tuned methods on web search and low-resource benchmarks.
Key insights
- An iterative, bi-directional refinement process between LLMs and RMs can mutually enhance their inputs, leading to improved information retrieval accuracy.
- Combining LLMs' natural language understanding with RMs' efficient access to vast, up-to-date data enables robust zero-shot dense retrieval without supervised training.
- The performance of such a synergistic system is significantly influenced by the capacity and quality of the large language model used for knowledge generation and refinement.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.