Drug adverse reactions (ADRs) is vital to minimize health risks and reduce drug development costs. Extracting ADRs (ADRE) from social media is a vital supplement to conventional pharmacovigilance databases. However, it remains particularly challenging due to the informal and noisy nature of usergenerated content. It combines the state space model with the convolutional module to efficiently capture long-range dependencies and extract more discriminative features, while completing the recognition and quantification of ADRs through a regression module. Experiments on the MedHelp Medical Forum dataset demonstrates that Mamba-ADR has a low computational complexity and an accuracy rate of 78.1%, and is superior to the state-of-the art methods. It provides a promising tool for ADRE in in drug discovery and development.
Zhang et al. (Thu,) studied this question.