Vaishali Kalra, Dr. Rashmi Agrawal, and Srishti Sharma


Vader lexicon, cross-domain sentiment analysis, sentiment analysis, hybrid model, supervised topic classification, domain adaptability


An increase in user data in social network websites has shown an increasing power of social networks for expressing user opinions on various domains which has turned online opinions into a very valuable asset. Traditional machine learning algorithms do not perform well for the ever-increasing domains for sentiment analysis when using the cross-domain method, in which the model is trained on one domain and evaluated on different domains. Hence, in this study, we describe a sentiment classification model that applies to any domain and uses both labelled and unlabelled data in the training process. The developed model is domain adaptable model which is a combination of lexicon-based and machine learning-based techniques. The proposed model also uses the topic classification approach where the six baseline topics are used for identifying and feeding the labelled data set for the targeted domain. The novelty of the proposed model is that the generated labelled data set of the given targeted domain can be used in the future for sentiment classification. This model is unlike deep learning models required a large amount of data set for the training purpose. On benchmark datasets, our proposed model outperforms several previous cross- domain models and achieves 83% classification accuracy [33].

Important Links:

Go Back