Existing information retrieval techniques are increasingly being used for ranking people on different platforms such as job portal websites or Freelance marketplaces. However, information retrieval methods have originally not been designed with that purpose in mind. For example, unlike in document retrieval settings, people ranking requires protection against discrimination of people based on sensitive attributes like gender, race, age, etc.
This project aims at the development of approaches for assessing and controlling bias in ranking people with the following four primary objectives.
- First, we intend to measure and model data biases present during the training of models.
- Second, we intend to measure and model user biases that manifest themselves when users interact with a ranking system via, for example, the provision of relevance feedback.
- Third, we intend to evaluate existing ranking algorithms and their ability to deal with data biases as well as user biases.
- Finally, we intend to synthesize our empirical findings into the development of approaches that are capable of generating bias-controlled rankings considering settings with or without feedback.
We believe our project will contribute towards the development of systems that adequately account for the complexities involved in ranking people. The methods and approaches developed in the project will also encourage further research in the direction of fair information retrieval.
The research project is based at the Chair of Data Science in the Economic and Social Sciences . In 2023, the German Research Foundation (DFG) provided funding of 313.296,00€ for a period of 36 months. The start of the research project is planned for 2024. Prof Markus Strohmaier is the contact person for inquiries.