Weakly supervised sentiment analysis using joint sentiment topic detection with bigrams

Online reviews evolve rapidly over time, which demands much more efficient and flexible algorithms for sentiment analysis than the current approaches. Current approaches detect the overall sentiment of a document, without performing an in-depth analysis to discover. We propose a Document level sentiment classification in conjunction with topic detection and topic sentiment analysis of bigrams simultaneously. This model is based on the weakly supervised Joint Sentiment-Topic model, and this extends the Latent Dirichlet Allocation by adding the sentiment layer. We considered Bigrams in ordered to increase the accuracy of sentiment analysis. We created a sentiment thesaurus with positive and negative lexicons and this is used to find the sentiment polarity of the bigrams. This model can be shifted to other domains. This is verified experimentally through four different domains which even outperforms the existing semi-supervised approaches.