Detecting Adverse Drug Reactions through Drug Common Sense Reasoning and Dual-channel Contrastive Learning

Adverse drug reactions (ADR) refer to harmful reactions that occur after patients take drugs under normal dosage and usage. Timely detection of potential adverse drug reactions can effectively reduce its negative impact on patients. The exponential growth of biomedical literature and the rapid development of social media have effectively promoted the development of natural language processing technology in the field of adverse drug reaction detection. The current mainstream approach is to use the pre-trained language model BERT for detection, but it lacks professional information about the biomedical field. At the same time, the vector generated by model does not have the ability to represent semantic relevance and is easily deceived by noise in social media. In this paper, we propose a Contrastive and Drug Commonsense-enriched ADR Detector (CDC-ADROR). An external knowledge module is added to the model, and the drug-related keywords in each sample are obtained from GPT-4 by using prompt learning, effectively solving the problem of BERT's lack of professional information. Meanwhile, contrastive learning is introduced to improve the loss function of the model, improving the quality of sentence representation vectors. Extensive experiments are carried out to evaluate the proposed framework on two datasets CADEC and Twitter.