A Content-based Reasoning Method for Multi-hop Question Answering using Graph Neural Networks

Abstract Question-answering systems require retrieving evidence from multiple documents or paragraphs and reasoning over them to meet users' information needs and answer their complex questions. On the other hand, the Explainability and comprehensibility of the predictions made by question-answering systems pose a challenge. In this paper, a content-based reasoning approach based on graph-based machine reading comprehension methods is proposed to answer multi-hop complex questions. In this approach, relevant paragraphs are selected in a two-step process after receiving the input of a multi-hop complex question. Then, to facilitate content-based reasoning and utilize the evidence related to the multi-hop complex question in the retrieved paragraphs, an incoherent graph infrastructure is constructed. Subsequently, a graph neural network and a transformer are employed as an encoder to extract the content-based answer relevant to the question from the graph infrastructure. Finally, to overcome the challenge of interpretability in the question-answering system, a transformer and the predicted answer are utilized. To evaluate the effectiveness of the proposed research, the proposed method is tested on the HotpotQA open-domain dataset with over 112,000 question samples. The results obtained in the relevant paragraph selection section show an improvement 0.58% in F1 compared to the best existing results. Furthermore, in the question answering section, improvements of 2.07%, 6.64%, and 9.97% are observed in F1, exact match and joint F1 metrics, respectively, compared to the best method. In the supporting fact prediction section, the results obtained in the proposed method demonstrate a significant superiority over the best existing works.