Aspect-Based Sentiment Analysis in Drug Reviews Based on Hybrid Feature Learning

This paper aims to develop a novel hybrid feature learning approach for aspect-based sentiment analysis to detect and classify the widely available unlabeled social data. The proposed approach employs Bidirectional Encoder Representations from Transformers (BERT) technique with Latent Dirichlet Allocation (LDA) model to predict the context words and learn the corresponding sentence and document vectors. Also, the proposed approach utilizes Bi-directional Long Short-Term Memory (Bi-LSTM) to classify the extracted sentiment. Various experiments are conducted on several social media datasets about adverse drug reactions (ADRs) reviews as a case study to evaluate the effectiveness of the proposed approach. Obtained experimental results show that the proposed feature learning approach outperforms other tested state-ofthe-art feature learning approaches and improves the feature and topic extraction for unstructured social media text and sentiment classification. The proposed approach achieved significant values of 95.4% for average accuracy, 0.935 for AUC score, and 94% for F-measure.