Is Graph Structure Necessary for Multi-hop Question Answering?
This challenges a trend in NLP by showing that graph-based methods may not be needed for multi-hop QA, potentially simplifying models for researchers and practitioners.
The paper investigates whether graph structures are necessary for multi-hop question answering, finding that with proper use of pre-trained models, graph structures can be replaced by self-attention or Transformers, as demonstrated on HotpotQA.
Recently, attempting to model texts as graph structure and introducing graph neural networks to deal with it has become a trend in many NLP research areas. In this paper, we investigate whether the graph structure is necessary for multi-hop question answering. Our analysis is centered on HotpotQA. We construct a strong baseline model to establish that, with the proper use of pre-trained models, graph structure may not be necessary for multi-hop question answering. We point out that both graph structure and adjacency matrix are task-related prior knowledge, and graph-attention can be considered as a special case of self-attention. Experiments and visualized analysis demonstrate that graph-attention or the entire graph structure can be replaced by self-attention or Transformers.