Trend Based Vertex Similarity for Academic Collaboration Recommendation

In this paper, we propose a new method used for collaboration recommendation in the academic domain. The proposed method is based on combination of probability theory and graph theory for modeling and analysing co-author network. In the co-author network, similar vertices are explored as potential candidates for collaboration recommendation. Taking the trend information into considering similarity of vertices in the network is the main contribution of this research. We did experiments with co-author networks extracted from the DBLP. Co-authorship that will occur in future used to evaluate accuracy of collaboration recommendation methods. We used metadata of publications from 2001 to 2005 for building the training network. The testing networks were built with publications from 2006-2008 (the testing network 1 for the near future prediction) and 2009-2011 (the testing network 2 for the far future prediction). The experimental results show that the proposed method, called TBRSS (Trend Based Relation Strength Similarity), outperforms other existing methods.