Multiple features-based adverse drug reaction detection from social media using deep convolutional neural networks (DCNN)

• Published in: Multimedia tools and applications 2024-01, Vol.83 (26), p.67779-67793
• Main Authors: Spandana, S., Prakash, R. Vijaya
• Format: Article
• Language: English
• Subjects: Accuracy; Artificial neural networks; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Digital media; Feature extraction; Flexibility; Multimedia Information Systems; Neural networks; Performance evaluation; Sensitivity analysis; Side effects; Social networks; Special Purpose and Application-Based Systems; State-of-the-art reviews
• ISSN: 1573-7721; 1380-7501; 1573-7721
• DOI: 10.1007/s11042-024-18144-9

Adverse drug responses (ADRs) are unfavourable side effects of using a medication that result from the medication's pharmacological activity. Social media has gained popularity recently as a forum for individuals to discuss their health issues. As a result, it is becoming a common place to get natural language information on ADR. Drug-related problems and side effects are crucial concerns in medical diagnosis. There is a higher likelihood of identifying the adverse effects of medications used for a particular ailment based on comments made on social media. Most works are developed based on conventional text features and a decision-based classification process. This work proposed to perform adverse drug reaction detection from social media-related comments. The convolutional neural network-based techniques are proposed further to increase the accuracy at higher levels. Conventional techniques can be performed with particular kinds of drugs for only some local regions worldwide. The flexibility of the technique is poor in the recently proposed conventional techniques. To overcome this, we propose a deep convolutional neural network-based ADR structure with diverse training for better flexibility and accuracy. Multiple features such as sentiment, statistical, and medical keywords-related features are extracted to perform an accurate training process to obtain highly accurate results. Various sensitivity evaluation metrics, such as accuracy, sensitivity, and specificity values, are evaluated to validate the performance of the proposed method. This method is compared with another state of the art to analyze the performance comparison.