Research map / Large language models / Instruction and fine-tuning
Long-Document QA with Chain-of-Structured-Thought and Fine-Tuned SLMs
Zhuowen Liang, Xiaotian Lin, Zhengxuan Zhang, Yuyu Luo, Haixun Wang, Nan Tang
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 |
|---|---|---|---|
| add process reward in RL (ablation) | LiteCoST (LLaMA) · Loong Finance | Average Score: 75.52 -> 76.95 | Improved |
| add process reward in RL (ablation) | LiteCoST (Qwen) · Loong Finance | Average Score: 77.39 -> 79.93 | Improved |
| add outcome reward in RL (ablation) | LiteCoST (LLaMA) · Loong Finance | Average Score: 72.55 -> 76.95 | Improved |
| add outcome reward in RL (ablation) | LiteCoST (Qwen) · Loong Finance | Average Score: 75.43 -> 79.93 | Improved |
| fine-tune LLaMA-3.2-3B with LiteCoST | LLaMA-3.2-3B-Instruct · Loong Finance | Average Score / Perfect Rate: 49.37 -> 76.95; 0.11 -> 0.40 | Improved |
| fine-tune Qwen2-7B with LiteCoST | Qwen2-7B-Instruct · Loong Finance | Average Score / Perfect Rate: 62.10 -> 79.93; 0.26 -> 0.48 | Improved |
| fine-tune LLaMA-3.2-3B with LiteCoST | LLaMA-3.2-3B-Instruct · Loong Legal | Average Score: 30.67 -> 45.45 | Improved |
| fine-tune Qwen2-7B with LiteCoST | Qwen2-7B-Instruct · Loong Legal | Average Score: 38.05 -> 44.94 | Improved |
| add CoST structured data before reasoning | GPT-4o · LongBench HotpotQA/2Wiki/NarrativeQA/Qasper | F1: 32.59 -> 35.09 NarQA; 67.75 -> 72.98 2Wiki | Improved |
| add CoST structured data before reasoning | Qwen2.5-14B-Instruct · LongBench | F1: 29.78 -> 31.77 NarQA; 60.00 -> 67.11 2Wiki | Improved |
| use Chain-of-Thought prompting instead of zero-shot | Qwen2.5-14B-Instruct · Loong Finance | Average Score: 75.60 -> 77.51 | Improved |
| use CoST-based LiteCoST instead of CoT | GPT-4o · Loong Finance | Average Score: 80.08 -> 82.39 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- StructRAG — StructRAG: Boosting Knowledge Intensive Reasoning of LLMs via Inference-time Hybrid Information Structurization (ICLR 2025) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- BABILong: Testing the Limits of LLMs with Long Context Reasoning-in-a-Haystack (NeurIPS 2024 · similarity 0.88)
- DocScope: Benchmarking Verifiable Reasoning for Trustworthy Long-Document Understanding (NeurIPS 2026 · similarity 0.87)
- TrustUQA: A Trustful Framework for Unified Structured Data Question Answering (AAAI 2025 · similarity 0.87)
- CofCA: A STEP-WISE Counterfactual Multi-hop QA benchmark (ICLR 2025 · similarity 0.87)
- Holistic Reasoning with Long-Context LMs: A Benchmark for Database Operations on Massive Textual Data (ICLR 2025 · similarity 0.86)
- UDA: A Benchmark Suite for Retrieval Augmented Generation in Real-World Document Analysis (NeurIPS 2024 · 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.
