Research map / Large language models / Retrieval and knowledge
QASA: Advanced Question Answering on Scientific Articles
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| remove evidential rationale generation (answer directly from selected paragraphs) (ablation) | FLAN-T5 (3B) full-stack QA · QASA full-stack QA | ROUGE-L: 19.32 -> 24.53 with rationale generation | Helped |
| finetuning FLAN-T5 on the QASA multi-step pipeline (vs InstructGPT pretrained) | full-stack QA · QASA full-stack QA | ROUGE-1: 32.22 vs 27.11 for InstructGPT | Helped |
| training on GPT-augmented synthetic data (ablation) | FLAN-T5 rationale generation · Rationale generation | ROUGE-1: 29.22 for GPT AUG-ONLY vs 27.63 for FLAN-T5 finetuned on all data | Helped |
| training only on ASQA data instead of combined data (ablation) | FLAN-T5 answer composition · Answer composition | ROUGE-1: 48.97 ASQA-only vs 45.59 combined | Mixed |
| finetuning on collected/combined training data (vs pretrained only) | FLAN-T5 associative selection · Associative selection | F1: 45.86 vs 34.72 for InstructGPT | Helped |
| using GALACTICA (scientific-paper pretrained) instead of general LMs | full-stack QA · QASA full-stack QA | ROUGE-1: 15.56 pretrained / 20.93 finetuned vs 22.48 / 32.22 FLAN-T5 | Hurt |
Closest papers by meaning
- Explore What LLM Does Not Know in Complex Question Answering (AAAI 2025 · similarity 0.87)
- CofCA: A STEP-WISE Counterfactual Multi-hop QA benchmark (ICLR 2025 · similarity 0.87)
- Tree-of-Reasoning Question Decomposition for Complex Question Answering with Large Language Models (AAAI 2024 · similarity 0.87)
- Thinker: Learning to Think Fast and Slow (NeurIPS 2025 · similarity 0.86)
- T-SciQ: Teaching Multimodal Chain-of-Thought Reasoning via Large Language Model Signals for Science Question Answering (AAAI 2024 · similarity 0.86)
- Mitigating Factual Hallucination in Large Reasoning Models via Mixed-Mode Advantage Regularization (NeurIPS 2026 · similarity 0.86)
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.