Weighted Content Based Methods for Recommending Connections in Online Social Networks

Online Social Networks currently have an important role in the life of millions of active internet users. Cases like Twitter are of special attention since a lot of connections are made between people who never met before and with no need of reciprocation. For this reason it is important to nd new ways to provide recommendations that may be of interest for users. Should these recommendations focus on the popularity, on the activity, location, common friends or content? Should recommendations be inuenced by egocentric or global network metrics? This research is the rst phase of an in-depth study of a large dataset based on Twitter which aims to answer the previous questions. Despite many studies based on global rankings, the authors believe that recommendations should mostly be based on the preferences made by users in their own networks. This stage of the study focuses on the popularity and activity of links as indicators to predict connections.For this end, the authors compute a weight for each of these features, which varies for each user. Each pair tested is accepted if it satises a minimum total weight. Results show a slight but important improvement in performance when using two features instead of one, the results gives an insight that if more features are considered more improvements in predictions will be found. The results of this paper can and should be accompanied with more research.