Some studies on cross domain sentiment analysis

Abstract

Recent technological developments in social media and the world newlinewide web have made people interact and express their attitudes and opinions newlinein various ways. These opinion based documents are rich resources, which are newlineintegral, subjective and valuable information available in huge amounts but newlinescattered in nature. These documents are needed by corporate, business newlinepeople, manufacturers, service providers and individuals to make the newlinedecision, to analyse the current trends and to understand the pulse or the newlinesocial sentiment of the public. Thus, sentimental analysis or opinion mining newlinewhich comprises information retrieval, text analysis, computational linguistics newlineand natural language processing systematically identifies, extracts, quantifies newlineand classifies subjective information of the online reviews. The sentiment newlineclassifier is domain-specific. A sentiment classifier for the hotel domain newlineclassifies sentiments with good accuracy within the domain only. This newlineresearch work aims to build a classifier that classifies sentiments of reviews newlinefrom different domains. newlineIt is very expensive and also requires exhaustive training to newlinegenerate a new classifier for every domain. Domain adaptation approaches newlinegenerally use labelled reviews for the training and testing process. Labelled newlinereviews are limited in nature. Unlabelled reviews are widely available but newlinediscriminating them is a tedious process. Polarity conflict is another issue newlinewhere a sentiment word would be positive sentiment in one domain but newlinenegative in another domain. Moreover, a sentiment classifier classifies newlinereviews into negative, positive and neutral. But certain application requires newlineclassification into various strength of opinions at different granularity levels. newlineAlgorithms and approaches have been used to overcome the above issues. newline