With 600,000 offers processed per month, France Travail is entering the era of artificial intelligence and launching LegO. Its goal is twofold: to eliminate non-compliant offers and to free up time for advisors.
Presented during the World AI Cannes Festival in February 2024, LegO is the name of the AI developed by Pôle Emploi, which became fully France Travail in February 2024, to validate the conformity of the job offers it publishes on its site web. This latest AI-powered tool analyzes the content of offers and verifies that they meet no fewer than 22 compliance criteria. LegO thus saves time and guarantees the quality of the offers offered to job seekers.
Improving the relevance of a time-consuming system
France Travail has 20 million job offers published on its site, 478 million visitors online, and 6.6 million in local agencies. Suffice it to say that the organization has all the makings of a multinational to run as efficiently as possible.
And in terms of tools for analyzing job offers, until now, it was necessary to rely on business rules that employees and advisors scrupulously applied to ensure that the advertisements published complied with French laws. Basically, apart from welcoming and providing support, they had to deal with
image classification, automation of CV analysis, email processing and even ensuring that job offers comply with legal standards.
But according to Agathe Ravilly, product manager within the AI department of France Travail, “ […] the proportion of false positives exceeded 30%, forcing advisors to spend time double-checking ads. Hence the idea of replacing the rules with AI to improve the relevance of the system “.
France Travail therefore decided to develop its own system. Called Intelligence Emploi, this solid program powered by AI is still in production.
Strong performance to ensure efficiency
It is within this program that LegO, the solution for analyzing job offers using artificial intelligence, was developed. The goal, according to Agathe Ravilly, being to “ replace a paid market solution, improve the quality of published job offers and reduce the time spent by advisors correcting job offers.”
And to do this, France Travail has provided the means.
Intelligence Emploi, an Open Source technology, runs on 28 GPUs with secure access to the in-house data lake. “ Our application runs on 6 LTSM (Long short-term memory) networks and performs 22 compliance checks », explains Agathe Ravilly.
On the security side, LegO is developed on-premise (everything is designed on site, to ensure optimal security) and open source. “ The platform also offers computing capabilities with CPU and GPU resources, resources which significantly reduce the time required to train complex models.continues Agathe Ravilly.
Finally, in terms of performance, LegO has it under the hood. Intelligence Emploi, which incubates LegO, runs on 28 GPUs and has secure access to the in-house data lake. Agathe Ravilly adds: “ Our application runs on 6 LTSM (Long short-term memory) networks », she concludes.
Encouraging results, but “can do better”
It was necessary to use 35 training datasets, select 50,000 offers, which experts analyzed, then marked (via an internal tool called Replay) to guarantee LegO foolproof precision.
The result was not long in coming with an accuracy rate of 82%, LegO makes it possible to analyze the conformity of hundreds of thousands of job offers proposed each month to France Travail before being put online.
And no less than 20% of them were designated by the AI as non-compliant, because they were discriminatory (prerequisites on personal property, compulsory driving license, possessions, requests based on gender, etc. .). “ We can say that it is a great success », rejoices Agathe Ravilly.
But if it remains very efficient, because it is available 100% of the time and has a response time of 300 milliseconds, LegO can also be improved, thanks to the hand of man. Indeed, advisors will have to manually correct the offers that LegO has mentioned as non-compliant, or adapt those that the AI has not correctly processed. They must also provide regular feedback to experts on the results obtained and their experience. Too much time is still spent by advisors assisting an AI rather than supporting job seekers.
The emergence of artificial intelligence as a mainstream tool has opened up numerous possibilities for all content producers. Text, image, sound… This new fashionable technology can now provide assistance in many areas, and facilitate work in the most difficult stages of creation. And with an ever-increasing offering, it is important to distinguish which tools provide real added value. So you don’t waste hours trying everything the Google results pages offer!
Read more
Source : The Digital Factory
3