A Co-authorship Based Random Walk Model For Academic Collaboration Recommendation

As a social network, the entity relationship of author collaboration network is that two author publish a paper collaboratively. Link prediction in social networks is an important issue, and its practical significance can be understood as recommending friend-nodes for nodes in a network. This thesis mainly makes targeted improvements for the author collaboration network based on the conventional Random Walk algorithm, and proposes a recommendation Random Walk algorithm based on link importance. The thesis also implements the recommendation system based on the improved algorithm and provides friend recommendation service.First, the coauthor recommendation based on author collaboration network is classified as link prediction problem of social networks in this thesis. In order to solve the sparse data problem in the author collaboration network, this thesis uses a recommendation algorithm based on Random Walk, and makes deeper research on the restarting Random Walk algorithm, whose classic representative is Google’s PageRank. Then, this thesis presents a recommendation Random Walk algorithm based on link importance. Link importance is defined in the thesis and the thesis gives three factors including the number of cooperation, time of each cooperating and the order of coauthors to build a mathematical model. The thesis also modifies the transition matrix of Random Walk so that the Random Walk in the migration process can have more tendency and guidance. Thus, the coauthor recommendation with improvements can be more precise either in theory or in reality.Finally, based on DBLP data sets, the thesis proposes the coauthor recommendation solution of author collaboration using the recommendation Random Walk algorithm based on link importance. What’s more, a web server is built using Redis database, and Tornado web framework to provide coauthor recommendation service based on author collaboration.