On the most representative summaries of network user activities

In our approach to summarizing user information, dubbed SUM, we develop two scoring mechanisms that cooperatively optimize for polarizing criteria. After mapping user activity information onto a category tree, the scoring mechanisms highlight the most representative tree node (or summary); the node provides an aggregated view of the activities most characteristic of the user. We evaluate our approach by using web activity on the network of a large Cellular Service Provider and summarizing it to devise interests of individual users as well as groups. We compare SUM against an algorithm that discovers Hierarchical Heavy Hitter and show that SUM uncovers previously unknown information about users.