Performance enhancement of machine Learning approaches for document level Sentiment classification

Abstract

In the present modern world, it has become essential for people to know othersand#8223; opinion before making any decisions. With the rapid advancement of web, people are interested to purchase the products and express their opinions online through review sites, blogs, discussion forums, social networking sites and so on. Due to the phenomenal growth of user-generated data on the web, it is time-consuming to manually read and analyze them to obtain some useful information. Hence, the necessity to automatically extract and analyze the peopleand#8223;s opinions has gained significance. Sentiment Analysis (SA) also called Opinion Mining (OM) is the study that analyzes peopleand#8223;s sentiments, opinions, emotions and evaluations expressed in written text. SA research can be classified into three levels of granularity namely document-level SA, sentence-level SA and feature-level or aspect-level SA. This research focuses on document-level sentiment analysis that classifies the entire review document as positive or negative. Semantic orientation approaches and machine learning approaches are used in the literature for SA. Semantic orientation approaches use a corpus or a dictionary to identify the polarity of the document. Machine learning approaches first build a model using the training data. The model built is then used to identify the class label of the unseen test data. Machine learning approaches have shown better performance for document-level sentiment classification. newlineResearchers have suggested various techniques to identify the sentiment of opinionated documents. Still, there are many issues that can be addressed to improve the performance of existing methods. newline