Cross-Domain Sentiment Analysis Employing Different Feature Selection and Classification Techniques

The paramount work of information mustering has been to find out what is the opinion of the people. Sentiment analysis is errand discerning the polarity for the given content which is dichotomized into two categories—positive and negative. Sentiment analysis operates on colossal feature sets of unique terms using bag of words (BOW) slant, in which case discrete attributes do not give factual information. This necessitates the elimination of extraneous and inconsequential terms from the feature set. Another challenging fact is most of the times, the training data might not be of the particular domain for which the perusal of test data is needed. This miscellany of challenges is unfolded by probing feature selection (FS) methods in cross-domain sentiment analysis. The boon of cross-domain and Feature Selection methods lies in significantly less computational power and time for processing. The informative features chosen are employed for training the classifier and investigating their execution for classification in terms of accuracy. Experimentation of FS methods (IG, GR, CHI, SAE) was performed on standard dataset viz. Amazon product review dataset and TripAdvisor dataset with NB, SVM, DT, and KNN classifiers. The paper works on different techniques by which cross-domain analysis vanquishes, despite the lower accuracy due to difference in domains, as better algorithmic efficient method.