Different paths to the same destination: Diversifying LLMs generation for multi-hop open-domain question answering
Abstract: Highlights•We propose DP-CoT, which utilizes two-level diverse CoT generation.•We design a BERT-style evidence classifier for reasoning path pruning.•The DP-CoT achieves 1.1%-4.8% improvement compared to GPT3-based baselines.
External IDs:dblp:journals/kbs/LiWWCM25
Loading