Application of a personalized collaborative filtering job recommender system for the Flemish employment service

Recommending vacancies to job seekers is a crucial part of an employment services' task to link supply and demand on the labor market. Knowledge based recommender systems based on user- and vacancy profiles are often applied in this context because of their high scrutability and safe recommendations. They do however require a large amount of knowledge engineering to set up. On top of this, our preliminary experiments show that there is often a mismatch between the type of jobs people explicitly indicate to be interested in, and the jobs the user's implicit feedback indicates an interest for. This causes recommendations coming from a knowledge based recommender system not always to be aligned with a user's interest. In this paper we propose and evaluate a collaborative filtering job recommender system based on implicit user feedback. Existing research shows that collaborative filtering recommender systems suffer from a cold-start problem and small neighborhoods of similar people in case of high sparseness, both of which also manifest themselves in our research. Besides these known challenges linked to collaborative filtering, we identify and propose solutions to new challenges that arise when collaborative filtering recommender systems are applied to job recommendation. The last part of this paper compares our collaborative filtering technique with the knowledge based recommender system as currently used by the Flemish employment services.status: publishe