Reproduce “CASTLE: Contrastive and Seed-Guided Training for Cold-Start Natural Language Search” 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
Wendy Ran Wei, Hao Li, Weiwei Guo, Xiaowei Liu, Xueyin Chen, Chun How Tan, Malay Haldar, Soumyadip Banerjee and 4 more
CASTLE generates synthetic queries and topical listing labels for a natural-language search mode without its own query history. It combines survey seeds with contrasting listings from booking sessions, prompting an LLM to distinguish each pair in a query. Survey-seed CASTLE’s query-length KL divergence is 1.01, versus 9.33 for Promptagator; CASTLE-R, using real-query seeds, scores lower at 0.75.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.