Deriving User's Profile from Sparse Egocentric Networks

Several studies demonstrate effectiveness and benefits of using user's social network information to enrich user's profile. In this context, one of our contributions [1] proposes an algorithm enabling to compute user's interests using information from egocentric network extracted communities. Therefore, mining information from a small or a sparse network remains challenging because there is not enough information to enrich a relevant user's profile. So, one of the main lock is to cope with the lack of information that is considered as an important issue to extract a relevant community and could lead to misinterpretations in the user's profile modeling process. We aim to improve the performance of [1], regarding the lack of information problem, in the case of a small and/or a sparse network. We propose to add more information (i.e. relations) into user's network before extracting the data and enriching his profile. To achieve this enrichment, we suggest using snowball-sampling technique to identify and add user's distance-2 neighbors (friends of a friend) into the user's egocentric network. Our experimentation conducted in DBLP demonstrates the interest of node integration into small and sparse network. This leads to the study of link prediction that enables us to provide better performances and results compared to the existing work.