Supervisor Recommendation In A Research Social Network

Supervisor selection is important for research students in their future studies and careers. Currently, students rely on information search or friends’recommendations to find potential research supervisors. However, due to the severe problems of incomplete and asymmetric information between students and supervisors, students can hardly find suitable supervisors that match their research interests as well as their personalities. Therefore, there is a pressing need to provide an intelligent supervisor recommendation approach which can support students to make reasonable decisions.In existing studies, various methods are proposed for handling the problem of supervisor selection. They mainly consider topic-relevance and candidate-quality, and often overlook the significance of connectivity consideration whose institutional-level connection can be mined through a collaboration network of published papers and whose individual-level connection can be extracted from a friend network on research-related social platforms. Extant approaches mention several one-sided criteria of personality requirements. However, they neglect the two-sided matching degree of the personality styles of the individuals involved.This thesis propose a novel supervisor recommendation approach that integrates the relevance, connectivity, quality and personality-matching dimensions in a research social network platform. The relevance analysis measures the similarity degree between a target student and potential supervisors in the academic information aspects. This dimension is computed by a method of discipline-supervised semantic relevance matching. The connectivity dimension combines the individual-level social connection and the institutional-level collaboration connection. In contrast to the traditional assessments of quality, the social popularity of a researcher is also considered in this work. Subsequently, a personality-matching aided measurement is employed to analyze the matching degree of two individuals’personality styles. The proposed approach is evaluated through a user study. Average rate and normalized discounted cumulative gain metrics are used to compare the results from different approaches. Experimental results show that our approach outperforms the baseline methods.The present solution has been implemented as a social network recommendation service on ScholarMate, which aims to connect people together to create innovation and conduct research smartly. On the one hand, the supervisor recommendation approach can support students in the process of making reasonable decisions on supervisor selections. On the other hand, it is a novel way of providing personalized dynamic knowledge resources for students, which can also potentially enhance their social learning opportunities.