Reproduce “Harnessing Large Language Models for Knowledge Graph Question Answering via Adaptive Multi-Aspect Retrieval-Augmentation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Derong Xu, Xinhang Li, Ziheng Zhang, Zhenxi Lin, Zhihong Zhu, Zhi Zheng, Xian Wu, Xiangyu Zhao and 2 more
The Adaptive Multi-Aspect Retrieval-augmented (AMAR) framework enhances Large Language Models (LLMs) for Knowledge Graph Question Answering (KGQA) by adaptively selecting and aligning multi-aspect knowledge. This system achieves state-of-the-art performance on WebQSP and CWQ datasets, demonstrating improved factual accuracy and logical form generation by mitigating noise from retrieved information.
Key insights
- Simply appending retrieved knowledge as context to LLMs significantly hinders performance due to noise and context length issues, while adaptive filtering and alignment are crucial.
- The self-alignment and relevance gating modules are critical for effectively leveraging multi-aspect knowledge, with ablation studies confirming their individual contributions.
- Relations play a particularly crucial role in logical form generation, as their removal led to the largest performance drop among knowledge aspects.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.