Research map / Large language models / Retrieval and knowledge
DeFuzzRAG: Handling Fuzzy Time Expressions for Temporal Robustness in Retrieval-Augmented Generation
Ling-Chun Chen, Hsi-Wen Chen, Ming-Syan Chen
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 temporal intent extraction, metadata inference and overlap filtering after dense retrieval | MPNet retriever · TempFuzz-QA fuzzy queries | Hit Rate: 0.6712 -> 0.7765 (LLaMA2-7B parser) | Improved |
| fuzzify temporal expressions in documents and queries | MPNet, BM25, BGE retrievers · TempFuzz-QA | Hit Rate: MPNet 0.7908 -> 0.6712; BM25 0.5574 -> 0.4587 | Hurt |
| fuzzify documents (temporal expressions made vague) | retriever attention · 1,000 query-document pairs | Time Attention Ratio: 27.7% average drop | Hurt |
| use a 7B parser (Mistral-7B) instead of GPT-4 for temporal parsing (ablation) | DeFuzzRAG · TempFuzz-QA fuzzy | MRR: 0.5483 -> 0.5402, latency 11.14 s -> 2.84 s | No clear effect |
| use TinyLLaMA-1.1B as temporal parser instead of 7B models (ablation) | DeFuzzRAG · TempFuzz-QA fuzzy | MRR: 0.5402 -> 0.4702 (vs Mistral-7B) | Hurt |
| add DeFuzzRAG on explicit (clean) queries | MPNet retriever · TempFuzz-QA explicit | Hit Rate: 0.7908 -> 0.8125 | Improved |
| add DeFuzzRAG to spatially fuzzified queries | MPNet retriever · SpaceFuzz-QA | Hit Rate: 0.7253 -> 0.7834 | Improved |
| use fuzzy type Generalized Dating instead of other fuzziness types (ablation) | DeFuzzRAG · TempFuzz-QA by fuzziness type | Hit Rate (fuzzy): GD lowest, about 0.57 | Hurt |
| query type with strong temporal anchors (Event-Anchored, Year-Specific) vs range queries (ablation) | DeFuzzRAG · TempFuzz-QA by query type | Hit Rate (fuzzy): ET about 0.90, YS about 0.87 | 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.
- MPNet — MPNet: Masked and Permuted Pre-training for Language Understanding (NeurIPS 2020) · better on 5 of 6 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
- Retrieval is Not Enough: Enhancing RAG through Test-Time Critique and Optimization (NeurIPS 2025 · similarity 0.88)
- Worse than Zero-shot? A Fact-Checking Dataset for Evaluating the Robustness of RAG Against Misleading Retrievals (NeurIPS 2025 · similarity 0.88)
- Long-Context LLMs Meet RAG: Overcoming Challenges for Long Inputs in RAG (ICLR 2025 · similarity 0.88)
- RAPID: Long-Context Inference with Retrieval-Augmented Speculative Decoding (ICML 2025 · similarity 0.87)
- Inference Scaling for Long-Context Retrieval Augmented Generation (ICLR 2025 · similarity 0.87)
- RAGGED: Towards Informed Design of Scalable and Stable RAG Systems (ICML 2025 · similarity 0.87)
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.
