Cross Domain Sentiment Classification by Extracting Best Opinion Features
Australian Journal of Basic and Applied Sciences, Vol. 10(2), Pages: 192-198, 2016
7 Pages Posted: 10 Jul 2017
Date Written: January 8, 2016
Background: Sentiment classification aims at classifying the reviews as positive or negative. Generally all learning techniques require labeled data but obtaining labeled data in every domain is impractical. However, applying trained classifier in one domain to another also produce poor performance due to different data distribution between the domains. In order to increase the classification performance in target domain, the framework has been proposed which select the best opinion features from both source and target domain when training a classifier in source domain. The proposed method select best opinion features by computing domain relevance score. In order to perform cross domain sentiment classification, sentimentally similar best opinion features of both source and target domains are identified by computing cosine similarity between the source and target domain features. Next, feature augmentation is performed while training a classifier in source domain. Finally, the classifier trained in source domain is applied to target domain to predict the sentiment of the reviews. Experiments are performed on Amazon product data sets by applying proposed methodology and experimental results show that best classification accuracy in target domain than the other state of art of previous approaches of cross domain sentiment classification.
Keywords: Data Mining, Cross Domain Sentiment Classification
Suggested Citation: Suggested Citation