Adverse drug reactions (ADR) have become a common and serious problem faced by drug users worldwide, posing a significant threat to human life and health safety. How to achieve automatic evaluation of the quality of ADR reports, and how to mine and evaluate the relevance between drugs and adverse drug reactions, has become an urgent problem that needs to be solved at present. In this study, a text classification technology based on deep learning were employed to establish an automated system to evaluate the information quality and relevance of ADR reports, using ADR reports from cooperative medical institutions and case studies in the literature as samples. The ERNIE+DCGNN model was used to train the ADR relevance evaluation model, and its effects were compared with other mainstream models. Comparative experimental results demonstrated that the ADR relevance evaluation model constructed in this paper had better experimental results.