HMM-based state prediction for Internet hot topic

In this paper, we build an on-line topic detection and state prediction system, which can automatically collect Internet web pages, cluster them into topics, and predict the hot topics´ states. A HMM-based prediction model is proposed to predict the Internet hot topic´s state, and the prediction method is testified in an actual network environment. In this system, we train the observations of the topics by the hidden Markov model and save the models in a HMM library for the topic´s prediction. Topics with similar life cycle are recorded and share a same model. Experimental results are shown.