Exploration of methodologies to improve job recommender systems on social networks

Abstract

This paper presents content-based recommender systems which propose relevant jobs to Facebook and LinkedIn users. These systems have been developed at Work4, the Global Leader in Social and Mobile Recruiting. The profile of a social network user contains two types of data: user data and user friend data; furthermore, the profile of our users and the description of our jobs consist of text fields. The first experiments suggest that to predict the interests of users for jobs using basic similarity measures together with data collected by Work4 can be improved upon. The next experiments then propose a method to estimate the importance of users’ and jobs’ different fields in the task of job recommendation; taking into account these weights allow us to significantly improve the recommendations. The third part of this paper analyzes social recommendation approaches, validating the suitability for job recommendations for Facebook and LinkedIn users. The last experiments focus on machine learning algorithms to improve the obtained results with basic similarity measures. Support vector machines (SVM) shows that supervised learning procedure increases the performance of our content-based recommender systems; it yields best results in terms of AUC in comparison with other investigated methodologies such as Matrix Factorization and Collaborative Topic Regression.