A hybrid approach for cold start recommendations

Abstract

The work presented in this paper is part of the system developed by our team for RecSys challenge 2017. XING is a social networking website that allows the recruiters to post their job openings. It also allows the interested candidates to look up these job openings and apply for the job if suits their requirements. The goal of the challenge was to identify user behavior for newly generated job postings; there are no past interactions available for these jobs. We used a hybrid approach to solve this problem: two ranked list were generated by two different systems and these lists were later merged to make the recommendations; one system is a memory based recommender system and the other is model based recommender system. We used an item based collaborative filtering method for memory based recommendations and XGBoost to train the model based system.