Hybrid Framework Integrating Lexicon and Learning Methods for Enhancing Sentiment Analysis Based on Patients' Tweets on Medicines

Abstract

- Sentiment analysis is the act of locating and categorizing the emotions conveyed in text data using text analysis tools. Previous studies have demonstrated the great potential for sentiment analysis of medicine reviews to provide useful data that will aid healthcare professionals and businesses in assessing the safety of pharmaceuticals after they have been sold. These details protect patients and strengthen their confidence in healthcare professionals. Existing frameworks for opinion investigation within the restorative field either take two types of methods one is lexicon methods and the other one is machine learning models. Learning-based approaches need annotated data, whereas lexicon-based approaches are domain-specific and have a smaller range of applications. To improve outcomes, this study employs a hybrid methodology that fusion machine and deep learning models with lexicon techniques. The reviews are annotated using all-purpose emotion lexicons like TextBlob and SenticNet. To extract meaningful features, we are using feature engineering methods TF and TF- IDF. Last but not least, classification tasks are carried out using learning models such as machine learning models and deep learning models for biomedical text. The execution of the prospective combined technique is appraised using performance metrics. According to experimental findings, combining lexicon- and learning-based techniques yields superior outcomes over using them alone. Additionally, TextBlob has demonstrated impressive results, providing an accuracy of 97% with LSTM-CNN and the Bio Bert model when used to a dataset of medication reviews, as well as 95% accuracy when used with TF and the logistic regression model. Additionally, TextBlob provides an accuracy of 94% when combined with TF and LSTM, and provides an accuracy of 97% when using with Bio Bert Model on a dataset including tweets.