Tie strength prediction in OSN

A decade earlier, work on modeling and analyzing social network, was primarily focused on manually collected dataseis where the friendship links were sparse but relatively noise free (i.e. all links represented strong physical relation). With the popularity of online social networks, the notion of “friendship” changed dramatically. The data collection, now although automated, contains dense friendship links but the links contain noisier information (i.e. some weaker relationships). The aim of this study is to identify these weaker links and suggest how these links (identification) play a vital role in improving social media design elements such as privacy control, detection of auto-bots, friend introductions, information prioritization and so on. The binary metric used so far for modeling links in social network (i.e. friends or not) is of little importance as it groups all our relatives, close friends and acquaintances in the same category. Therefore a popular notion of tie-strength has been incorporated for modeling links. In this paper, a predictive model is presented that helps evaluate tie-strength for each link in network based on transactional features (e.g. communication, file transfer, photos). The model predicts tie strength with 76.4% efficiency. This work also suggests that important link properties manifest similarly across different social media sites.