Hybrid One-Class Collaborative Filtering for Job Recommendation

Intelligent recommendation has been a crucial component in various real-world applications. In job recommendation area, developing an effective and personalized recommendation approach will be very helpful for the job seekers. In order to deliver more accurate job recommendations, we propose a system by incorporating users’ interactions and impressions as the data source and design a hybrid strategy by taking the advantages of three existing one-class collaborative filtering (OCCF) algorithms. The proposed solution combines multi-threading techniques in the traditional item-oriented OCCF (IOCCF) and useroriented OCCF (UOCCF) algorithms, and also applies an approximation of the sigmoid function in Bayesian personalized ranking (BPR) to improve the efficiency of the overall performance. Based on the experiment results, the proposed system shows the effectiveness of using users’ interactions and impressions by an improvement of 23.78%, 15.61% and 11.50%, and the effectiveness of using the hybrid strategy by a further improvement of 34.55%, 16.20% and 20.90%, over IOCCF, UOCCF and BPR, respectively.