Clustering Social Networks Using Interaction Semantics and Sentics Praphul Chandra, HP Labs India, praphul.chandra@hp.com Erik Cambria, National University of Singapore, cambria@nus.edu.sg Amir Hussain, University of Stirling, ahu@cs.stir.ac.uk Abstract The passage from a static read-only Web to a dynamic read-write Web gave birth to a huge amount of online social networks with the ultimate goal of making communication easier between people with common interests. Unlike real world social networks, however, online social groups tend to form for extremely varied and multi-faceted reasons. This makes very difficult to group members of the same social network in subsets in a way that certain types of contents are shared with just certain types of friends. Moreover, such a task is usually too tedious to be performed manually and too complex to be performed automatically. In this work, we propose a new approach for automatically clustering social networks, which exploits interaction semantics and sentics, that is, the conceptual and affective information associated with the interactive behavior of online social network members. 1. Introduction Online social network representations often aggregate a user’s social network into a common cluster of ‘friends’. This approach is acceptable when the context of interaction is specific (e.g., LinkedIn [1]) but can lead to problems when the interaction context is broad and generic (e.g., Facebook [2]). The problem is further complicated by the fact that it is not always easy to demarcate the interaction context into ‘specific’ or ‘generic’. Often, the interaction context is an emergent property of how users come to use the medium and different users choose to use it in different ways, e.g., some users may find it acceptable to add a family member as a LinkedIn contact whereas others may not. Even when tools are available for users to be able to classify their friends into different clusters, they are rarely used [3]. Previous user studies have attributed this to (a) this process being time consuming and tedious; (b) the dynamic nature of the groups where individuals need to be added or removed from the group on a case-by-case basis [4]. This motivates the automatic clustering of a user’s social network, which is updated dynamically based on user’s social interactions. For clustering social networks, we must differentiate between a user’s ego-centric networks and a global socio-centric networks. Whereas, a user’s ego-centric network only contains ‘friends’ of a particular user, the global socio-centric network is formed by combining the ego-centric networks of all users in the system. In this work, we focus initially on a user’s ego-centric network. Later we show, how our approach may be extended to socio-centric networks. An automatic dynamic clustering of users’ personal social network is useful to enable users to segregate their online social networks and reflect more accurately social networks in the real world. This is crucial, e.g., for meeting the privacy expectations of users (preventing unintended sharing). Clustering a user’s social network is also useful for recommending potential group members during group communication and sharing – this is especially useful on mobile devices where interaction space is limited [3].