Research map / Large language models / Alignment and preferences
Breaking the Reference Bottleneck via Learning to Rewrite Conversational Queries without Gold Reference Passages
Doyoung Kim, Youngjun Lee, Joeun Kim, Jihwan Bang, Hwanjun Song, Susik Yoon, Jae-Gil Lee
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use CQR model itself (dual role) for response refinement instead of Llama+ICL (Optimized) (ablation) | DualReform · SciConvQA (sparse retriever) | retrieval MRR: 19.11 -> 20.06 | Improved |
| use CQR model (dual role) instead of plain Llama for response refinement (ablation) | DualReform · SciConvQA (sparse retriever) | pseudo reference MRR: 36.55 -> 50.05 | Improved |
| add in-context learning to single-role Llama refiner (ablation) | Llama refiner · SciConvQA (sparse retriever) | pseudo reference MRR: 36.55 -> 44.82 | Improved |
| add iterative optimization to single-role Llama+ICL (ablation) | Llama+ICL · SciConvQA (sparse retriever) | retrieval MRR: 19.02 -> 19.11 | No clear effect |
| increase pseudo reference updates from 1 to 3 (ablation) | DualReform · SciConvQA (sparse retriever) | retrieval MRR: 17.55 -> 20.06 | Improved |
| increase pseudo reference updates from 1 to 3 (ablation) | DualReform · TopiOCQA (sparse retriever) | retrieval MRR: 25.87 -> 28.39 | Improved |
| increase pseudo reference updates from 2 to 3 (ablation) | DualReform · TopiOCQA (sparse retriever) | retrieval MRR: 28.08 -> 28.39 | Improved |
| add query-forming Prompt 1 template (ablation) | DualReform · SciConvQA (sparse retriever) | pseudo reference MRR: 45.38 -> 50.05 | Improved |
| add query-forming Prompt 1 template (ablation) | DualReform · TopiOCQA (sparse retriever) | pseudo reference MRR: 54.57 -> 56.50 | Improved |
| include response text ({at}) in Prompt 1 (ablation) | DualReform · TopiOCQA (sparse retriever) | pseudo reference MRR: 50.00 -> 56.50 | Improved |
| include phrase state the main points in Prompt 1 (ablation) | DualReform · SciConvQA (sparse retriever) | pseudo reference MRR: 43.62 -> 50.05 | Improved |
| include phrase last response {at} in Prompt 1 (ablation) | DualReform · TopiOCQA (sparse retriever) | pseudo reference MRR: 47.16 -> 56.50 | Improved |
Closest papers by meaning
- Response Enhanced Semi-supervised Dialogue Query Generation (AAAI 2024 · similarity 0.87)
- MaFeRw: Query Rewriting with Multi-Aspect Feedbacks for Retrieval-Augmented Large Language Models (AAAI 2025 · similarity 0.86)
- Learning to Clarify: Multi-turn Conversations with Action-Based Contrastive Self-Training (ICLR 2025 · similarity 0.86)
- ESPRESSO: An Effective Approach to Passage Retrieval for High-Quality Conversational Recommender Systems (AAAI 2025 · similarity 0.86)
- ChatQA: Surpassing GPT-4 on Conversational QA and RAG (NeurIPS 2024 · similarity 0.85)
- Making Retrieval-Augmented Language Models Robust to Irrelevant Context (ICLR 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
