Sentiment Analysis using Feature Generation And Machine Learning Approach

The study of opinion offers answers to what the most critical problems are. Since sentiment analysis can be automated, judgements can be taken based on a significant amount of data rather than plain intuition, which is not always accurate. This paper focuses on the feature generation using Bag-of-Words and TF-IDF and the build model using the machine learning approach for sentiment analysis. The dataset used contains review of trip advisor on various hotels. This dataset consists of 20k reviews. Word cloud had been formed using sentiment ratings. Data was cleaned and pre-processed, and then applied Bow and TF-IDF for feature extraction. After implementation of classifiers, training and evaluation was performed. Evaluation metrics is used for measuring the accuracy of classifier. MultinomialNB obtained the highest accuracy in the realm of Bag of Word features and random forest outperformed in the case of TF-IDF out of three classifiers used to determine accuracy. We got 82% of the classification rate of MultinomialNB in Bag of Word and 78% accuracy in TF-IDF Random Forest.