Reviews posted on online in general require more efficient and flexible algorithms for sentiment analysis than the methods that are currently available approaches. Overall sentiment of a document is detected without performing a deep analysis in most of the approaches currently available. So the result obtained for the accuracy of classification is not good. In order to increase the accuracy of classification, topic detection is combined with document level sentiment classification. Bigrams are considered to give better classification. Latent Dirichtlet Allocation method is used with Joint Sentiment Topic detection for classification in weakly supervised sets. In the proposed method this is used with naive bayes algorithm to further improve the accuracy of classification. Using Latent Dirichlet Allocation method, a sentiment thesaurus is created with positive and negative lexicons to find the sentiment polarity of the bigrams. Three different domains are experimented with the new method and it is found that the proposed model can also be applied to any other domain. From the results obtained it could be inferred that the proposed methodology gives better accuracy in classification than the existing semi-supervised approaches