Collaborator Recommendation System Based On Coauthor Network

This paper is mainly about studying and building a collaborator recommendation system based on coauthor network. Main contents include three parts:the building of coauthor network, the calculation of recommendation similarity, the design and implement of collaborator recommendation system.1. The building of coauthor network. The publications meta-data for collaborator system is collected from web crawler and user upload. For constructing the coauthor graph efficient and dynamic, retrieve and store the coauthor relation static, then update it locally when the publications dataset changed.2. The calculation of recommendation system. Considering the criteria of collaborator recommendation system, the recommendation similarity includes two aspects:one is vertex similarity and the other is topic similarity:(l)Vertex similarity calculation, this paper propose a method named weighted-path relation strength(WPRS) similarity. By experiment, we proved that WPRS is more efficient and reasonable in the real application. Also, we can use the breadth-first search to initialize all the vertex similarity in acceptable time cost. (2) Topic similarity calculation, this paper use the Latent Dirichlet Allocation(LDA) to clustering the paper abstracts to infer the topic of this paper, and save the trained model for inferring the coming paper. Use the metric named Significance Of Interest(SOr) as the calculation of topic similarity. At last, combine two similarity above using the weighted exponential function, we can get the final similarity of recommendation system.3. Design and implement of collaborator recommendation system. The whole system contains four functional modules:coauthor-graph searching module, collaborator recommendation module, user authority module and the publications upload module.