A Survey on Feature Selection Methods in Sentiment Analysis

Customers give their opinions in the form of positive, negative, or neutral comments about a service or product on social media platforms like Twitter, Internet Movie Database (IMDb), and e-commerce Web sites like Amazon. People use these opinions to make decisions about whether to purchase products and services or not. However, for further analysis, data must extract. Therefore, sentiment analysis is employed to collect opinion and other data and change it into useful information. Sentiment analysis (SA) is the process of analyzing the sentiment of a text in the form of positive or negative. SA performed on these opinions may give incorrect results when irrelevant, redundant, and high-dimensional features are selected. Feature selection (FS) is a SA preprocessing step that improves classification accuracy by reducing feature dimensionality and overcoming underfitting and overfitting by selecting relevant features. FS also reduces computation time, complexity, and storage. This paper reviews a comparative study of various filter and wrapper methods used by researchers to select the best features. This study includes datasets, evaluation metrics, FS and extraction methods, classifiers, results, and shortcomings of various methods. This study has proved that the filter method has less accuracy while the wrapper method has a high computational cost. The research’s findings demonstrate that hybrid FS methods remove the drawbacks of filter and wrapper approaches.