Impulse-Response Model for Human Behaviors Sequences

• Published in: 2015 IEEE International Conference on Data Mining Workshop (ICDMW) pp. 1040 - 1047
• Main Authors: Li, Houyi, Han, Banghe, Li, Ying, Li, Junming
• Format: Conference Proceeding Journal Article
• Language: English
• Published: IEEE 01.11.2015
• Subjects: Behaviors Prediction; Conferences; Data models; Encyclopedias; Heavy Tail; Human; Human behavior; Human Dynamics; Internet; Intervals; Mathematical model; Mathematical models; Predictive models; Recommendation Algorithm; Reproduction; Statistics; Time factors; Timing
• ISSN: 2375-9259
• DOI: 10.1109/ICDMW.2015.116

The interval time distribution is a well investigated in the area of 'human dynamic'.Many research explained the heavy tail phenomenon and reproduced the heavy-tail-like interval time or response time distribution with various models. This paper empirically studies human online activities both at individual level and group level based on 'T-mall' data set and 'Wikipedia' data set. It points out that the statistic features of human behaviors with acquainted objects and unacquainted objects need to be considered independently. Based on research in these two data sets, the timing of human behaviors is a combination of the heavy tail distribution for time interval of executing acquainted objects and the quasi uniform distribution for initial time of executing unacquainted objects. It's shown that this phenomenon is a consequence of inherent causality within human behaviors. This paper proposes Impulse-Response Model to describe this causality. This model connect the two famous problem in human behavior research: the reproduction problem and prediction problem. Time interval distribution of T-mall data set is well reproduced by this model. This paper also show that Impulse-Response Model hold a higher accuracy to make prediction about human future behaviors than traditional classifications both in T-mall data set and Wikipedia data set.