Scientific Users' Interest Detection and Collaborators Recommendation

With the development of science and technology, the academic research is continuously evolving and the subject category is getting more and more refined. Meanwhile, as the internet information technology develops, the communication between researchers is getting closer. Detecting scientific users' interest and finding collaborators with common research interest to them are not only good ways to promote academic cooperation, but also effective personalized service methods to increase scientific users' satisfaction, for instance, recommending review experts and building virtual research teams. In this paper, we propose a scientific user interest detection method that combines cross-collaboration-domain features based on the LDA (Latent Dirichlet Allocation) model, and present the collaborators and the article recommendation algorithms based on it. To further improve the algorithm efficiency for the user interest detection, we construct the distributed LDA model under Hadoop Map-Reduce framework. Experiments on a large scale of Wanfang library and NNSF (National Natural Science Foundation) of China illustrate that, our proposed recommendation algorithm outperforms the collaborative filtering algorithm and the single interest model recommendation algorithm.