Keycoopt and Inria bring machine learning to the world of recruitment by referral
Keycoopt's concept is to enable companies to recruit managers through employee referrals. Launched in 2012, this Lille-based start-up called upon Inria's expertise to design a software tool in order to automate and reinforce the relevance of recruitment via employee referral.
The recruitment market is becoming increasingly digitalized through professional social networks (Linkedin, Viadeo, etc.). However, networking also plays an essential role: according to the latest APEC (French agency for the employment of managers) annual survey, 39% of companies say that they have turned to referrals by staff members in order to find candidates. Keycoopt, a young Lille-based start-up, proposes linking these two trends using an "e-referral" platform. The principle behind it? It sends targeted job advertisements, on behalf of companies, to a network of business experts (almost 30,000 in 2016) based on their locality, profession and sector. Named "co-opters", they are then tasked with mobilizing their network of contacts in order to recommend the candidate whose profile best matches the recruiter's requirements. Keycoopt enables companies to meet candidates who are hard to identify using traditional tools. "Through referrals, we can unearth an atypical profile who makes all the difference and who would never have applied! ” Keycoopt points out.
An algorithm to optimize the selection of co-opters
In May 2015, Keycoopt signed a partnership contract with the Magnet research team (jointly with the University of Lille 3*) from Inria Lille - Nord Europe, specialized in the definition of "machine learning" or "artificial learning" methods and models within information networks. The aim of this partnership: to automate the co-opter selection system, which until now was carried out "manually" by Keycoopt staff. François Noyez, an engineer dedicated to technology transfer within the InriaTech program (see below ) and Rémi Gilleron, IT professor and Magnet researcher, worked together in order to propose a software solution. The research team therefore performed a "matching" between the non-textual data in the job advertisements (classification by category, sector, department, geographical area, profession type...) and those belonging to the co-opters, in order to identify those best-suited to propose relevant profiles. The classification algorithm uses wide-margin linear separators. It was set up as a prototype in Python language. "It is a programming language that is widely used within the research community, for which there are already effective machine learning modules and toolboxes ", states François Noyez. "The program was then integrated within the Keycoopt information system ".
"Machine learning" at the service of "crowd recruiting"
Just like KissKissBankBank, numerous start-ups and digital companies call on Keycoopt services. Of the 600 companies convinced by this recruitment solution, there are also some big names such as L'Oréal, Orange or Cartier. The result: increased reactivity and responsibility for a wider range of referral opportunities."That is where machine learning comes in: based on quantitatively significant data, it makes it possible to find correlations that humans cannot find alone or not as easily ", Rémi Gilleron points out. "In fine, predictions are more pertinent, and correspond more to reality ", underlines the data scientist François Noyez.
The automatic learning software designed for Keycoopt, which was delivered in December 2015, could be enriched and become the subject of a new partnership over the coming months. "Refine the relevance of the automatic referrals would be a question of making use of the co-opters' textual data, such as career path and other elements from their curriculum vitae, and to cross them with data focusing more on the full description of the job advertisement as provided by the Keycoopt customer " François Noyez explains. "Moreover, not only use the fact of knowing if the person was hired in the end or not, but also the more precise reasons behind this decision, using comments made by the co-opters and Keycoopt consultants ", Rémi Gilleron adds. This would also enable co-opters to receive job advertisements that are increasingly in keeping with their professional network and, for Keycoopt, to reduce the duration of the search for referable candidates.
Machine learning: the recruiter of tomorrow?
Machine learning used in referrals constitutes a new research and development issue for Inria, which benefits from the technological expertise provided by the recent InriaTech initiative. "Scientifically, this research and development project provides us with the opportunity to explore new text and network analytical methods ", Rémi Gilleron is delighted to say. Keycoopt, which positions itself as the TripAdvisor of recruitment, is in line with new collaborative economy companies. Even so, is "crowd recruiting" the sign of an "uberisation" of recruitment? Can machine learning eventually replace a recruiter or a human resources officer? "For the time being, the aim is more modest: it is more a question of replacing the head-hunter model by collaborative methods, which automatic learning can help to recommend, suggest, and select in very large solution spaces. It is a scientific challenge that meets the needs and current issues of our society, which is increasingly turned towards the collaborative economy", underlines Rémi Gilleron. "The aim here is to put forward proposals that are subject to validation by a business expert. In my view, if the machine can propose profiles, the recruitment decision remains a human matter. ", François Noyez concludes.
InriaTech: an initiative to facilitate technology transfer
InriaTech, which was put in place in April 2015, is an initiative of the Inria Lille - Nord Europe center. It is dedicated to technology transfer and the development of R&D in the Lille area. The idea behind this initiative is to further partnerships between Inria project teams and innovative companies from the Lille area. Based at EuraTechnologies, InriaTech currently brings together around ten engineers in charge of projects and partnerships with companies. As such, each collaborator has his or her field of expertise: man-machine interface, machine learning, robotics, statistical meshing... "This structure represents an important evolution for Inria.It makes it possible to react to short-term collaborative projects with companies, and to spread the circle of software development expertise to new areas, such as data science, interactive software, or robotics ", François Noyez describes.
Its mission is also to encourage and facilitate the creation of start-ups coming from Inria project teams. Funded primarily by the Hauts-de-France regional council, the European Metropolis of Lille (Métropole Européenne de Lille) and supported by The European Regional Development Fund (ERDF), InriaTech hopes to become self-sufficient within the next three years through the funding of research and development projects by industry partners.
* within the UMR 9189, CNRS-Centrale Lille-Université Lille1, CRIStAL.
These articles could interest you:
For more information
Inria / Photo H. Raguet