Dynamic Topic Detection Model by Fusing Sentiment Polarity

Traditional static topic models mainly focus on the statistical correlation between words, but ignore the sentiment tendency and the temporal properties which may have great effects on topic detection results. This paper proposed an LDA-based dynamic sentiment-topic (DST) model, which could not only detect and track topics but could also analyse the shift of general’s sentiment tendency towards certain topic. This model combines the data with the sentiment and dynamic properties of time by maximum likelihood estimation and the sliding window. We use Gibbs sampling method to estimate and update model parameters, and use random EM algorithm for model reasoning. Experiments on real dataset demonstrate that DST model outperforms the existing algorithms. .