For sentiment analysis of web short texts

a topic sentiment combining model (TSCM) is proposed based on LDA and web review behavioral theory

which is founded on the assumption that topic distribution of each sentence in a review is unique and different from that of other sentences.Generative process of TSCM is to first determine sentiment orientation of each word and then topic of each sentence in a review while taking word relation into consideration.Extensive experiments on real-world datasets (Movie and Amazon) show that TSCM significantly outperforms JST

S-LDA

D-PLDA and SAS in terms of the accuracy of sentiment classification and topic detection.