Interpretable multi-hop knowledge reasoning for gastrointestinal disease

Gastrointestinal diseases cause a large number of outpatient treatments every year and cost a lot of medical and health budgets. This continuous rise in healthcare costs, along with the need for improved diagnostic and treatment strategies, has encouraged researchers to investigate advanced methods to address the issue. In this paper, we propose an Interpretable Multi-hop Knowledge Reasoning (IMKR) method for gastrointestinal diseases and design an intelligent question-answering system framework. Our proposed method is rooted in the Dual Process Theory (DPT), which argues that reasoning involves two processes, including intuitive and fast process, and reflective and comprehensive process. Drawing inspiration from DPT, we use the Tensorlog operator and recursive neural network to generate logic rules corresponding to the first process of DPT, and further introduce cognitive expansion and subgraph reasoning to enhance the performance of IMKR, which achieves the second process. Moreover, to better evaluate the predictions of one-to-many relationships in the medical context, we also propose a new model evaluation algorithm. Finally, we conduct experiments on the healthcare platform dataset and the proposed method achieves the performance of hit rate of ranking 1 (Hits@1) 0.9542, hit rate of ranking 10 (Hits@10) 0.9995, and Mean Reciprocal Ranking (MRR) 0.6976.