Drug review sentiment analysis using bidirectional Lstm for medication recommendation

Abstract

The increasing pressure for affordable healthcare, particularly amid the coronavirus pandemic, has heightened the demand for smart drug recommendation systems. A lot of people turn to self-medication out of limited exposure to medical doctors, which subsequently results in harmful health consequences. The proliferation of online health communities and drug rating websites has posed a challenge of utilizing patient opinion for sentiment analysis and drug suggestion. This work introduces a new method that makes use of sentiment analysis of patient reviews to suggest suitable medications. This work suggests a sentiment analysis framework consisting of the blend of Logistic Regression, Naïve Bayes, and a Recurrent Neural Network (RNN) built on a Bidirectional Long Short-Term Memory (BiLSTM) network for drug review classification and the suggestion of medications from past symptoms and experiences. It incorporates several deep learning-based natural language processing (NLP) algorithms such as vectorization, Bag of Words (BoW), TF-IDF, Word2Vec, and Manual Feature Analysis, to retrieve meaningful information from drug reviews and make conditionbased drug recommendations. Machine learning classifiers such as LinearSVC are employed to predict sentiment and evaluate drug efficacy. Precision, recall, and F1score, accuracy, and AUC score confirm the efficacy of the system, with TF-IDF vectorization being used by the LinearSVC classifier to have 93% accuracy.