Research on Event Prediction Algorithm Based on Event Sequence Semantic

Abstract

Event prediction in event stream is an important problem in temporal data mining. However, existing event prediction algorithms are based on string prediction in which a character represents an event or an event type, do not take into account event sequence semantic and can not predict for infrequent event sequences. In this paper, an event prediction algorithm based on event sequence semantic called SVClustering-SVR is proposed to predict probability of target event occurrence in event stream in appointed interval. We build a vector structure called semantic vector to express event sequence semantic, and then utilize the attributes of standardizing semantic vector and confidence of rule which is generated by event sequences and target event to form samples space. Finally, we use Support Vector Regression (SVR) to build prediction model. To improve the accuracy of prediction, we also define semantic distance between event sequences and cluster semantic vectors. SVClustering-SVR algorithm can predict for infrequent event sequences and those not appeared in training set. Experimental results show the effectiveness of SVClustering-SVR algorithm.