Personalized community detection in scholarly network

Most graph clustering methods partition a network into communities based solely on the topology and structure of the network. Due to this, the means via which communities are detected on a network are insensitive to the preferences of a user who is searching the network with a specific, personalized information need. Such partition algorithms may be of diminished value for scholars exploring networks of research if these scholars possess prior preferences on what information they consider relevant. To better address this type of information seeking behavior, we introduce a personalized community detection algorithm that provides higher-resolution partitioning of areas of the network that are more relevant to a provided seed query. This algorithm utilizes the divisive Girvan-Newman approach but incorporates a user’s personal preferences as a prior. We show that this personalized algorithm can produce a more fine-tuned partition of a scholarly network when compared to existing prior-insensitive approaches.