Predictability of action time for online human behaviors

Abstract

With the widespread use of internet technology, the online behaviors become a more and more important part in human's daily lives. Knowing the time of user's next action in online activities is quite valuable for improving online services, which prompts us to wonder whether the time of user's next online activity is predictable? In this paper, we study the predictability of action time for human online activities using the dataset from a social network. To this end, we map the inter-event time sequence of user's online activities to a sequence of inter-event time symbols and analyze it using information-theoretic method. Results show that knowing the time interval between the current activity and previous activity decreases the entropy about the time interval between the next activity and current activity, i.e., in the inter-event time sequence, the knowledge of an inter-event time can help decrease the entropy about its next one, which indicates that the time of next online activity is predictable. Moreover, the short and long inter-event times decrease the entropy about the next inter-event time more largely than the medium ones, which indicates that the short and long inter-event times have higher predictive powers. Furthermore, our results show that the action time of online activities in weekdays is more predictable than that in weekends.