Analysis of feature extraction techniques for sentiment analysis of tweets

Over the past few years, sentiment analysis has moved from social networking services like LinkedIn, Facebook, YouTube, Twitter, and online product-based reviews to determine public opinion or emotion using social media textual contents. The methodology includes data selection, text pre-processing, feature extraction, classification model, and result analysis. Text pre-processing is an important stage in structuring data for improved performance of our methodology. The feature extraction technique (FET) is a crucial step in sentiment analysis as it is difficult to obtain effective and useful information from highly unstructured social media data. A number of feature extraction techniques are available to extract useful features. In this work, popular feature extraction techniques including bag of words (BOW), term frequency and inverse document frequency (TF-IDF), and Word2vec are compared and analyzed for the sentiment analysis of social media contents. A method is proposed for processing text data from social media networks for sentiment analysis that uses support vector machine as a classifier. The experiments are carried on three datasets of different context namely US Airline, Movie Review, and News from Twitter. The results show that TF-IDF consistently outperformed other techniques with best accuracy of 82.33%, 92.31%, and 99.10% for Airline, Movie Review, and News datasets respectively. It is also found that the proposed method performed better than some existing methods.