Mitigating Gender Bias in Job Ranking Systems Using Job Advertisement Neutrality

Transformer-based Job Ranking Systems(JRSs) are vulnerable to societal biases inher-ited in unbalanced datasets. These biases oftenmanifest as unjust job rankings, particularlydisadvantaging candidates of different genders.Most bias mitigation techniques leverage can-didates’ gender and align gender distributionswithin the embeddings of JRSs to mitigate bias.While such methods effectively align distribu-tional properties and make JRSs agnostic togender, they frequently fall short in addressingempirical fairness metrics, such as the perfor-mance gap across genders. In this study, weshift our attention from candidate gender tomitigate bias based on gendered language injob advertisements. We propose a novel neu-trality score based on automatically discoveredbiased words in job ads and use it to re-rankthe model’s decisions. We evaluate our methodby comparing it with different bias mitigationstrategies and empirically demonstrate that ourproposed method not only improves fairnessbut can also enhance the model’s performance.