Live mentoring services are recent novel social media, in which mentees can input expertise requests and wait for accepting some expert mentor who is willing to tackle the requests in a live and one-by-one manner. While mentee’s satisfaction of being mentored is determined by the matched mentor, it is crucial to have an effective mentor–mentee matching. This paper aims at recommending mentors based on the requests in Codementor, which is one of the popular live mentoring services. An accurate mentor recommendation will support the mentees’ decisions in finding suitable mentors, support the mentors’ decisions in filtering out irrelevant requests, and support the mentoring services’ decisions in assigning mentors to mentees. We divide the mentor recommendation problem into two tasks, Mentor Willingness Prediction (MWP) and Mentee Acceptance Prediction (MAP). MWP is to predict whether a mentor is willing to tackle a request, while MAP is to predict whether a mentee user will accept a recommended mentor. We propose to simultaneously deal with such two tasks by recommending a ranked list of mentors such that the recommended mentors who are really willing to tackle the request are as many as possible (MWP) and the final mentor who is accepted by the mentee can be ranked as high as possible (MAP). We develop four categories of features, availability, capability, activity, and proximity, to model the willingness of a mentor dealing with the request and the potential of a mentee to accept the recommended mentor. By applying various supervised learning methods, experimental results show the effectiveness of these features and provide extensive analyses to reveal more factors that can affect the quality of mentor recommendation. In addition, we also conduct a user study on Codementor platform to exhibit the practical performance of the proposed method. The innovation of this work includes the formulation of MWP and MAP problem in online mentoring services, feature engineering for mentoring prediction tasks, and data-driven experimental studies in prediction and a practical user study.
All Science Journal Classification (ASJC) codes
- Information Systems
- Human-Computer Interaction
- Hardware and Architecture
- Artificial Intelligence