A Weighted Variable Order Markov Model for Event Sequences
WU Hong-h · Jisuanji gongcheng · 2014
Variable-order Markov Model(VLMM) is a simple but effective model for event sequences modeling. However, the classic VLMM only considers the transition probability, without taking into account the frequency of the substring. A Weighted VLMM(WVLMM) is proposed in this paper, constructing a Weighted Probabilistic Suffix Tree(WPST) via the frequency of the substring based on the classic VLMM. It also proposes a strategy for branches pruning based on the degree of the similarity of the nodes while constructing the tree, in order to improve the generalization ability of the model, and to construct the tree in a linear time complexity. To validate the effectiveness of the model, the proposed model is applied to the classification of event sequences. Experimental results demonstrate that the new model can make an effective classification on real-world sequence datasets in different applications.