Skip to Main Content
Sentiment prediction for text has been an intriguing subject for the last few years. The goal of it is to automatically indicate the positive or negative attitude towards a topic of interest. The proliferation of user generated content on the World Wide Web has made it possible to perform large scale mining of public opinion. This paper presents an original implementation of a system that integrates a recently proposed semi-supervised learning algorithm for text polarity classification. Lexical prior knowledge is harnessed in conjunction with labeled and unlabeled documents. The presented method is based on joint sentiment analysis of documents and words and uses a bipartite graph representation of the data. Our system is integrated into Rapid Miner, which does not come yet with semi-supervised learners.