New retrieval method CoDeR improves accuracy for constrained queries

Researchers have introduced CoDeR, a novel method for information retrieval that addresses the limitations of relying solely on semantic similarity, particularly for queries involving constraints like negation or exclusion. CoDeR employs a dual-encoder approach, maintaining a topical encoder for broad candidate coverage and adding a separate compatibility scorer. This scorer is trained using contrastive learning on satisfying and violating evidence pairs, enabling it to distinguish between documents that are topically relevant but contradict the query's constraints. The system can then re-rank candidates or retrieve an auxiliary set, improving retrieval accuracy without requiring large language models at inference time. AI

RANK_REASON This is a research paper describing a new method for information retrieval. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.IR (Information Retrieval) →

• arXiv
• large language model

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →