Characterizing Cross-Domain Search Behavior

The heterogeneous nature of information needs and information sources requires techniques that efficiently combine and leverage insights from diverse data sources. Moreover, user interaction from different sources provides different perspectives on user’s interests and preferences. In this work, we consider user interaction data from different verticals (news, search) and characterize behavioral differences among users. Traditional research on crossdomain methods has focused on leveraging insights for the same user from different domains. Instead, in this work, we highlight the need to consider user groups based on the cross-domain information and show that users from these groups behave differently. We investigate a number of search characteristics including re-querying behavior, topical spread of user interests and the overall popularity of queries across the different user groups, and demonstrate how considering different user groups has implications for evaluating and designing cross-domain personalization and recommendation approaches.