Matching Jobs and Resumes: a Deep Collaborative Filtering Task

Thomas Schmitt (1, 2) , Philippe Caillou (3, 2) , Michèle Sebag (3, 2)

1 CNRS - Centre National de la Recherche Scientifique
2 TAO - Machine Learning and Optimisation
3 LRI - Laboratoire de Recherche en Informatique

This paper tackles the automatic matching of job seekers and recruiters, based on the logs of a recruitment agency (CVs, job announcements and application clicks). Preliminary experiments reveal that good recommendation performances in collaborative filtering mode (emitting recommendations for a known recruiter using the click history) co-exist with poor performances in cold start mode (emitting recommendations based on the job announcement only). A tentative interpretation for these results is proposed, claiming that job seekers and recruiters − whose mother tongue is French − yet do not speak the same language. As first contribution, this paper shows that the information inferred from their interactions differs from the information contained in the CVs and job announcements. The second contribution is the hybrid system Majore (MAtching JObs and REsumes), where a deep neural net is trained to match the collaborative filtering representation properties. The experimental validation demonstrates Majore merits, with good matching performances in cold start mode.

Mots clés: job recommendations; collaborative filtering; hybrid collaborative filtering; cold start; human resources; job announcements; resumes

Domaines: Informatique [cs]

Thomas Schmitt, Philippe Caillou, Michèle Sebag. Matching Jobs and Resumes: a Deep Collaborative Filtering Task. GCAI 2016 - 2nd Global Conference on Artificial Intelligence, Sep 2016, Berlin, Germany. ⟨hal-01378589⟩