Classical-quantum hybrid transfer learning for adverse drug reaction detection from social media posts

Adverse Drug Reactions (ADRs) is a threat to human beings, sometimes it causes death. Thus, the detection of ADRs is crucial. This paper introduces a classical-quantum hybrid deep learning model for detecting adverse drug effects from social media reviews (in English). The suggested methodology improvises the ability to grasp human language for identifying pharmacological side effects of a drug. This article aims to combine the features of traditional machine learning methods to quantum computing into a single framework. The knowledge gained from the pre-trained classical model is transferred to quantum model, improving the performance of quantum algorithms. The Variational Quantum Circuit (VQC) encodes, classical data and classifies input reviews into feature vectors. It creates a relationship network for each drug, based on condition and review comments, enabling class wise prediction to detect ADRs in a drug. The accuracy of quantum inspired transfer learning based proposed model architecture is quite encouraging and provides a good enough solution. In identifying ADRs from the study of reviews, this work obtained accuracy of 97% with a training loss of 0.0659 and validation loss of 0.072. A Bio-BERT based quantum transfer hybrid methodology is proposed for detecting adverse drug reactions from social media reviews. It provides more accurate prediction on ADRs of drug rather than earlier introduced methodologies. The proposed methodology emphasizes towards the quantum enhanced hybrid model design in research.