New framework CLEAR improves conversational retrieval by focusing on answerability

Researchers have developed CLEAR, a novel framework designed to improve conversational retrieval systems by focusing on answerability rather than just topical relevance. This framework utilizes entailment distillation to train a reranker to distinguish passages that support answers from those that are merely topically related. CLEAR also incorporates an abductive recall module that uses LLMs to infer potentially answerable queries from passages, thereby expanding the candidate pool. Experiments on datasets like TopiOCQA, QReCC, and TREC CAsT show CLEAR consistently outperforms existing baselines, particularly in conversations with significant topical noise. AI

IMPACT Enhances conversational AI systems by improving the accuracy of information retrieval, particularly in noisy contexts.

RANK_REASON Academic paper detailing a new framework for conversational retrieval. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.IR (Information Retrieval) →

• alphaXiv
• arXiv
• CatalyzeX
• CLEAR
• CORE Recommender
• DagsHub
• Gotit.pub
• Hugging Face
• Influence Flower
• QReCC
• ScienceCast
• TREC CAsT

• paper
• other

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