Email recipient prediction using reverse chronologically arranged implicit groups

Abstract

Although social networking has significantly influenced online communication, email still has managed to retain its importance. There are number of techniques proposed in past by researchers for recipient prediction/suggestion. Most of them are complex to implement and takes good amount of computation time. The major factor behind higher time complexity and space complexity is the prediction models these methods use. These days mobile device applications are being widely used for emailing and thus appropriate techniques should be found considering constraints of mobile devices. Keeping this in view our research focuses on proposing prediction model, which takes very less computational efforts to be maintained. Apart from this, existing methods focus on maximizing number of intended recipients in one prediction cycle. In this paper, we also propose a different way of looking at the problem, by targeting 1 intended recipient in each iteration. For this, we introduce hit rate as a good measurement technique to measure the effectiveness of recipient prediction algorithm. We also present a flaw in the compiled version of Enron data set, and show some novel analysis on Enron data set which will help immensely in creating efficient recipient prediction algorithm.