Methods for Optimal Feature Selection for Sentiment Analysis

Abstract

AbstractRecently, the research on sentiment analysis has raised enormously. In sentiment analysis, the opinions of the user are analyzed. The opinions given by the user on any product, movie, place, or food are generally in text form which is then eradicated to identify the positive, neutral, and negative opinions. Customer reviews provide a significant amount of data each day, thus the features that can be drawn from this data are very valuable. An efficient feature selection method is required which generated features that can be separated from each other and has minimum co-relation. In this paper, we have used four feature selection methods, namely forward selection wrapper method, decision tree method, chi-square method, and ANOVA method for selecting the optimal features from the given data. Logistic regression, support vector machine, random forest, Naive Bayes, K-nearest neighbors, and k-means classifiers are examined to see how feature selection approaches affect their performance.KeywordsFeature selection techniquesMachine learningSentiment classificationSpam reviews