Research
Experimental methodology
-
An Adaptive Experiment to Boost Online Skill Signaling and Visibility
Digital matching platforms promise to reduce frictions on the labor market by providing low-cost information on available positions and candidates. As such, they may form a welcome addition to the toolbox available to Public Employment Services to bridge labor supply and demand. However, there are certain challenges associated with their adoption. For instance, vulnerable populations may face difficulties in utilizing digital tools effectively. In this study, we evaluate the impact of a communication campaign by email designed to encourage the use of an online matching platform maintained by the French Public Employment Service, Pôle emploi. We designed several email templates that combined information, support or motivational content to encourage jobseekers to engage with their profiles on the platform. In order to discover email effectiveness we implement an adaptive experiment (contextual bandit) where the goal is to use past jobseekers take-up responses and characteristics to determine email allocation in the future, reducing gradually the allocation of less promising templates. Additionally, we built an optimal personalization allocation strategy based on collected data and test its effectiveness. Emails had a positive impact on the usage of the platform, as measured by a wide range of outcomes. However, attempts at learning a personalized emailing strategy did not manage to significantly improve on a random allocation of email templates.
Draft
Labor Markets
-
(Mis)Perceptions of the Labor Market
Accepted at NBER Conference, 2026
Worker investments and job search are shaped by perceived market returns, which depend on the preferences of firms and competing workers. Similarly, firm compensation and amenity provision depend on beliefs about the preferences of workers and competing firms. We study whether employers and workers hold accurate perceptions about the preferences of others in the labor market through a series of incentivized field experiments on one of Europe’s largest freelance platform. First, we elicit employer and worker preferences over a rich set of worker, employer, and job characteristics using randomized worker profile and job offer evaluations. Second, we elicit beliefs about these preferences, from the other side of the market as well as from competitors, and we test their accuracy. We document substantial perception gaps about labor market preferences on both sides of the market, including about the value of posted wages, human capital, market reputation, job amenities, and demographic characteristics. Third, we implement a personalized information treatment to correct these gaps and show that it changes participants’ market behavior. To help interpret our results, we develop a theoretical framework showing how the perception gaps we document can distort both pricing and investment decisions, reducing matching and welfare. Our findings highlight an important source of information frictions in labor markets and their economic consequences.
Forthcoming -
From Hidden Skills to Opportunities : The Impact of Job Portals use on Labor Market Outcomes
In this paper, we propose to investigate the benefits of the use of digital matching platform. In partnership with the French Public Employment Service, Pôle emploi, we encourage randomly selected registered jobseekers to fill-in and publish their profile on the PES platform making them more visible to recruiters. We are able to track the effects of the intervention on a large variety of outcomes, including employment, through rich administrative data and web logs provided by the French Employment Service
Work In Progress
AI Ethics & Recommender Systems
-
Evaluating LLM Behavior in Hiring: Implicit Weights, Fairness Across Groups, and Alignment with Human Preferences
Published: January 16, 2026
This study examines how general-purpose LLMs assign importance to different criteria in recruitment contexts. Using synthetic datasets based on real freelancer profiles and project descriptions from a European marketplace, we apply factorial experimental designs to measure how models weight various hiring-relevant factors. Key findings reveal that while LLMs prioritize productivity signals like skills and experience, they also interpret certain features beyond their explicit matching purposes. The study shows minimal average discrimination against minority groups while identifying intersectional patterns where productivity signals carry varying importance across demographic categories.
Keywords: Large Language Models, Person-job Fit, Fairness, Interpretability
Malt Research Paper -
What Do AI Recruiters Value? Auditing the Implicit Hiring Criteria of Large Language Models
Large Language Models are increasingly integrated into recruitment systems, yet their implicit hiring criteria remain largely unexplored. This study uses conjoint experimental designs to systematically audit what candidate attributes LLMs value when evaluating hiring decisions. Building a synthetic benchmark based on real-world freelance recruitment scenarios, we estimate the weights multiple models (Google, Anthropic, Mistral, DeepSeek) assign to skills, experience, education, rates, reputation, communication style, and demographic characteristics. Preliminary results reveal substantial heterogeneity across models: while all prioritize productivity signals, the relative importance of credentials, pricing, and demographic attributes differs markedly. The paper argues that selecting an AI hiring system is not merely a technical decision but a normative one, as different models encode distinct assumptions about candidate quality that may reshape access to employment and labor market outcomes.
Work In Progress -
Toward Job Recommendation for All
IJCAI (AI And Social Good Track), 2023. Also presented at ECML PKDD, AI4HR Workshop, 2023
This paper presents a job recommendation algorithm designed and validated in the context of the French Public Employment Service. The challenges, owing to the confidential data policy, are related with the extreme sparsity of the interaction matrix and the mandatory scalability of the algorithm, aimed to deliver recommendations to millions of job seekers in quasi real- time, considering hundreds of thousands of job ads. The experimental validation of the approach shows similar or better performances than the state of the art in terms of recall, with a gain in inference time of 2 orders of magnitude.
Proceedings -
Fairness in Job Recommendations: Estimating, Explaining, and Reducing Gender Gaps
AEQUITAS Workshop @ ECAI 2023
Algorithmic recommendations of job ads have the potential to reduce frictional unemployment, but raise concerns about fairness due to biases in past data. Our research investigates the issue of algorithmic fairness with a specific focus on gender in a hybrid job recommendation system developed in partnership with the French Public Employment Service (PES), which is trained on past hires. First, by viewing job ads as a set of characteristics (such as wage and contract type), we document how the algorithm treats job seekers differently based on gender, both unconditionally and conditionally on their search parameters and qualifications. Second, we discuss the notion(s) of algorithmic fairness applicable in this context and the trade-offs involved. We show that the considered system reflects some existing differences in hiring or applications but does not exacerbate them. Finally, we consider adversarial de-biasing technique as a practical tool to demonstrate the trade-offs between recall and reduced differentiated treatment.
Proceedings