Peer recommendation in dynamic attributed graphs

Abstract

In this paper, we propose a novel model to recommend possible research peers to a user efficiently. We model all the authors along with their attributes as an attributed graph and then perform community search on this attributed graph to find the most appropriate peer for a user. We propose algorithms to recommend a dynamic attributed graph efficiently and create a real-time community search model by deploying an incremental training algorithm. We also propose dynamic weighted attributes for each node (peer). Given a node and set of attributes (query), the proposed model is capable of self-expansion of the attribute set leading to a more significant match between two nodes. Our experimental results show that our proposed model achieved substantial performance gains over the existing models. In a nutshell, we present an intelligent system that incorporates the changes going on in the research world and suggest up-to-date recommendations. The proposed model is also robust enough to ensure that the recommendations do not suffer due of poor queries. © 2018 Elsevier Ltd