Connecting Users with Similar Interests across Multiple Web Services

Most online social networking services provide a feature for users to build interest groups. Based on the profiles and be- havior data, web services can assist users to join groups by recommending relevant interest groups. In this paper, we pro- pose a novel method to connect users across multiple services based on user-labeled tags. Tags represent interests of a user and have advantages in terms of privacy, up-to-datedness, and service coverage. We have collected tags from six popu- lar web services, and analyzed usage patterns. We observe that the popularity of tags is highly skewed and dependent on the web services. We have also found that a set of tags of a single user frequently changes over time. Through user study, we demonstrate that the vector space model combined with intra-personomy normalization is good enough to rec- ommend other users with similar interests.