Reducing Gender Bias in LinkedIn Job Ads: The Impact of Automatic Rewriting on Candidate Diversity

Details

Supervisors

Vande Kerckhove, Corentin

Faculty

Louvain School of Management

Degree label

Master [120] : ingénieur de gestion, à finalité spécialisée

Abstract

enThis thesis explores gender bias in job advertisements posted on LinkedIn, proposing an automated method to mitigate this bias. The twofold objective is to detect linguistic bias in job advertisements and to rewrite the texts automatically to make them more neutral, while measuring the impact on application diversity. To achieve this, a composite gender bias score is developed that combines lexical and vector approaches. The ads are then reworded using an autoencoder model designed to neutralise bias while preserving the original meaning. The effectiveness of this approach is evaluated by simulating LinkedIn's recommendation system. The results demonstrate that the rewritten advertisements continue to be offered to relevant candidates but reach a more gender-balanced audience, thereby mitigating the exclusionary impact of language. This work opens up promising prospects for more equitable recruitment by intervening at an early stage in the drafting of job advertisements. However, it also highlights that the quality of automatic rewrites needs to be improved significantly, as some phrases still sound unnatural or lack meaning. This underlines the need for further research into rewriting methods that are both more fluid and grammatically reliable.