Enhancing message collaboration through predictive modeling of user behavior

Abstract

Research studies have shown that the effectiveness of collaboration and the choice of communication modality is intricately linked with the perceived presence and availability of the collaborating parties. Most collaboration systems offer users the ability to publish their presence for effective collaboration. However, a close observation of users' behavioral data shows a divergence such as in a published `busy' state a user is actually willing to collaborate with certain people or in a published `available' state a user is unwilling to collaborate with certain people. This behavior makes the notion of presence in collaboration systems ineffectual and often unreliable. In this paper, we propose a new predictive model of behavioral presence for collaborative messaging systems that automatically infers multiple presence states based on users expected collaboration behavior towards a contact. We present a novel confirmatory data mining technique that overlays a `cluster of interest' on standard clustering techniques such as k-means, fuzzy k-means, and consensus clustering. We present validation results of our predictive model on data obtained from real-world deployed enterprise servers across multiple locations over a period of seven months.