Reproduce “BASES: Large-scale Web Search User Simulation with Large Language Model based Agents” 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
Ruiyang Ren, Peng Qiu, Yingqi Qu, Jing Liu, Wayne Xin Zhao, Hua Wu, Ji-Rong Wen, Haifeng Wang
A framework leveraging LLM-based agents, BASES, was developed to simulate large-scale web search user behaviors, leading to the creation of the WARRIORS dataset. This research provides a scalable and privacy-preserving method for generating diverse, personalized user interaction data that enhances information retrieval model performance.
Key insights
- LLM agents can effectively mimic human search behavior, demonstrating high consistency in query generation and rewriting, and largely realistic click patterns that align with user profiles.
- Synthetic data generated by BASES consistently improves information retrieval model performance, enabling superior results for tasks like session search and click prediction with substantially less training data.
- The framework produces distinct, personalized search behaviors for individual users and provides a valuable resource for low-resource scenarios by augmenting training data, enhancing model robustness.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.