Machine Learning Approach for Negation Handling in Sentiment Analysis

With the growing popularity of online shopping, product reviews have emerged as significant source of product information. Sentiment analysis was developed as a result of the growing customer interest in reading specific product reviews before making a purchase, which uses natural language processing to examine client opinions. The influence of negation on word scope and order, as well as the inability to handle linguistic features, is an essential task in sentiment analysis. In this study, we first tackle the challenge of assigning the polarity of a phrase while identifying the scope of negation. Second, we suggest an approach to dealing with negations that relies on linguistic features to identify the impact of various genres of negation in lexicon- dependent on sentiment analysis. The data imbalance problem was overcome using SMOTE. The results indicate that the hybrid approach, which includes a lexicon-based dictionary for training the machine learning classifier, performs better at detecting negative sentences. When comparing classifiers, Random Forest outperformed Support Vector Machine and Naive Bayes. However, Naive Bayes (NB) outperforms in terms of computing time.