Research map / Large language models / Retrieval and knowledge
DPRM: A Dual Implicit Process Reward Model in Multi-Hop Question Answering
Xinyi Wang, Yiping Song, Zhiliang Tian, Bo Liu, Tingjin Luo, Minlie Huang
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 co-training of CoT-PRM and KG-PRM (ablation) | DPRM · CWQ | F1: 0.5419 -> 0.5988 | Improved |
| add co-training of CoT-PRM and KG-PRM (ablation) | DPRM · CWQ | Hit@1: 0.6824 -> 0.7301 | Improved |
| add co-training of CoT-PRM and KG-PRM (ablation) | DPRM · WebQSP | F1: 0.7009 -> 0.7228 | Improved |
| add iterative reasoning (ablation) | DPRM · CWQ | F1: 0.5824 -> 0.5988 | Improved |
| add iterative reasoning (ablation) | DPRM · WebQSP | F1: 0.7095 -> 0.7228 | Improved |
| add iterative reasoning (ablation) | DPRM · WebQSP | Hit@1: 0.8587 -> 0.8715 | Improved |
| add both co-training and iteration (ablation) | DPRM · CWQ | Hit@1: 0.6692 -> 0.7301 | Improved |
| use KG-PRM instead of ImplicitPRM-DPO as reward model | Llama-3.1-8B reasoning · CWQ | F1: 0.3538 -> 0.4283 | Improved |
| use KG-PRM instead of ImplicitPRM-DPO as reward model | Llama-3.1-8B reasoning · CWQ | Hit@1: 0.5650 -> 0.6341 | Improved |
| use KG-PRM instead of ImplicitPRM-DPO as reward model | Llama-3.1-8B reasoning · WebQSP | Hit@1: 0.7266 -> 0.7346 | Improved |
| train PRMs on 1,000 pairs instead of full data | DPRM · WebQSP | F1: 0.7228 -> 0.6670 | Hurt |
| train PRMs on 1,000 pairs instead of full data | DPRM · CWQ | Hit@1: 0.7301 -> 0.7190 | Hurt |
Closest papers by meaning
- Tree-of-Reasoning Question Decomposition for Complex Question Answering with Large Language Models (AAAI 2024 · similarity 0.89)
- Linking Process to Outcome: Conditional Reward Modeling for LLM Reasoning (ICLR 2026 · similarity 0.88)
- Fixing the Broken Compass: Diagnosing and Improving Inference-Time Reward Modeling (ICLR 2026 · similarity 0.87)
- Rethinking Reward Models for Multi-Domain Test-Time Scaling (NeurIPS 2026 · similarity 0.87)
- Question Calibration and Multi-Hop Modeling for Temporal Question Answering (AAAI 2024 · similarity 0.87)
- Knowledge Editing with Dynamic Knowledge Graphs for Multi-Hop Question Answering (AAAI 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.
