,
Department of Computer Science and Engineering, Jain Deemed to be University, Bengaluru, Karnataka, India
,
Department of Computer Science and Engineering, Jain Deemed-to-be-University, Bengaluru, Karnataka, India
Department of Computer Science and Engineering, Jain Deemed-to-be-University, Bengaluru, Karnataka, India
Social Media Detection of Adverse Drug Reactions (ADRs) has evolved as an important technique for today's pharmacovigilance. Real-time detection through social media platforms such as Twitter has been proven effective but has been hindered by issues related to linguistic noise, syntactic variations, and imbalanced data. In this paper, a new Deep Convolutional Recurrent State Space Model (DCR-SSM) coupled with a decision tree (DT) and a bag of words (BoW) is presented. Domain-specific normalization and filtering of noise will be conducted to improve data quality. Feature engineering techniques include high-impact feature selection using the Decision Tree approach with information gain and BoW encoding of features in numeric vector format. DCR-SSM is designed to exploit the benefits of the convolution operation for feature extraction, recurrent neural networks with Bi-LSTM/GRU layers for sequence processing, and State Space Models (SSM) for managing long-distance dependencies. The experimental tests carried out using the SMM4H and TwiMed datasets prove that the architecture is superior, registering an accuracy score of 93.81%, a precision rate of 91.33%, a recall score of 96.80%, and an F1 score of 93.99%. In addition, the architecture scored a commendable ROC-AUC score of 97.90%. The results confirm that the architecture performs effectively when handling complex social media datasets.
This is an open access article distributed under the Creative Commons Attribution Non-Commercial License (CC BY-NC) License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
The statements, opinions and data contained in the journal are solely those of the individual authors and contributors and not of the publisher and the editor(s). We stay neutral with regard to jurisdictional claims in published maps and institutional affiliations.
