Adverse Drug Reaction Detection From Social Media Based on Quantum Bi LSTM With Attention

Drug combination is very common in the course of disease treatment. However, it inevitably increases the overall risk of adverse drug reactions (ADRs). It is very important to early and accurately detect and identify the potential ADRs for combined medication safety and public health. Social media is an important pharmacovigilance data source for ADR detection. But the data are complex, mass, clutter, highly sparse, so it is difficult to detect the ADR information from these data. Deep learning stands out in terms of increased accuracy. However, it takes a lot of training time and requires a lot of computing power. Quantum computing has strong parallel computing capability, and requires less computing power. By introducing attention mechanism and quantum computing into Bi-directional Long Short-Term Memory (Bi-LSTM), a quantum Bi-LSTM with attention (QBi-LSTMA) model is constructed for ADR detection from social media big data. QBi-LSTMA is composed of 6 variable component subcircuits (VQC) stacked. Under the condition that the main topology of Bi-LSTM remains unchanged, the biases of QBi-LSTMA in input gate, forgetting gate, candidate memory unit and output gate are removed to simplify the network structure, and the weight and active value qubits of the model are used to update the network weight. The performance of the proposed method is evaluated on the SMM4H dataset, comparing with one traditional ADR detection method and three deep learning based ADR detection approaches. The experiment results show that the proposed method has great potential in ADR detection.