Research map / Large language models / Retrieval and knowledge
HalluClean: A Unified Framework to Combat Hallucinations in LLMs
Yaxin Zhao, Yu Zhang
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 structured plan-then-reason detection to task routing (ablation) | GPT-3.5-turbo · QA detection | F1: 39.3% -> 67.8% | Improved |
| add task-oriented routing prompt to direct ask (ablation) | GPT-3.5-turbo · Summarization detection | F1: 24.7% -> 60.0% | Improved |
| add task-oriented routing to direct ask (ablation) | GPT-3.5-turbo · MWPs detection | accuracy: 59.5% -> 59.5% | No clear effect |
| add structured reasoning after routing (ablation) | GPT-3.5-turbo · Dialogue detection | accuracy: 67.8% -> 69.3% | Improved |
| HalluClean vs direct ask | GPT-3.5-turbo · Self-contradiction detection | accuracy: 64.0% -> 87.0% | Improved |
| HalluClean vs ChatProtect on dialogue | GPT-3.5-turbo · Dialogue detection | F1 and accuracy: F1 72.0% -> 74.3%; accuracy 69.3% -> 69.3% | Mixed |
| switch backbone to DeepSeek-V3 | HalluClean · Math word problems detection | accuracy: 81.5% -> 89.5% | Improved |
| add retrieval augmentation | HalluClean · HaluEval QA detection | F1: 67.8% -> 80.4% | Improved |
| apply HalluClean to Chinese benchmark | GPT-3.5-turbo · HalluQA | accuracy: 46.5% -> 55.0% (HalluQA); 50.0% -> 51.5% (CMHE-HD) | Mixed |
| apply HalluClean revision module | GPT-3.5-turbo · QA revision | hallucination reduction rate: 20.5% -> 72.5% | Improved |
| HalluClean revision on DeepSeek-V3 vs direct ask | DeepSeek-V3 · QA revision | hallucination reduction rate: 49.0% -> 74.0% | Improved |
Closest papers by meaning
- FactTest: Factuality Testing in Large Language Models with Finite-Sample and Distribution-Free Guarantees (ICML 2025 · similarity 0.93)
- ANAH-v2: Scaling Analytical Hallucination Annotation of Large Language Models (NeurIPS 2024 · similarity 0.92)
- LLM-Check: Investigating Detection of Hallucinations in Large Language Models (NeurIPS 2024 · similarity 0.92)
- Scalable Token-Level Hallucination Detection in Large Language Models (NeurIPS 2026 · similarity 0.92)
- Joint Evaluation of Answer and Reasoning Consistency for Hallucination Detection in Large Reasoning Models (AAAI 2026 · similarity 0.92)
- Teaching Language Models to Hallucinate Less with Synthetic Tasks (ICLR 2024 · similarity 0.92)
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.
