Microblog Hot Spot Mining Based on PAM Probabilistic Topic Model

Chongqing University, Chongqing, China

Abstract

Microblogs are short texts carried with limited information, which will increase the difficulty of topic mining. This paper proposes the use of PAM (Pachinko Allocation Model) probabilistic topic model to extract the generative model of text’s implicit theme for microblog hot spot mining. First, three categories of microblog and the main contribution of this paper are illustrated. Second, for there are four topic models which are respectively explained, the PAM model is introduced in detail in terms of how to generate a document, the accuracy of document classification and the topic correlation in PAM. Finally, MapReduce is described. For the number of microblogs is huge as well as the number of contactors, the totally number of words is relatively small. With MapReduce, microblogs data are split by contactor, document-topic count matrix and contactor-topic count matrix can be locally stored while the word-topic count matrix must be globally stored. Thus, the hot spot mining can be achieved on the basis of PAM probabilistic topic model.