Skip to main navigationSkip to main content
The University of Southampton
Web Science Institute

Observing and Recommending from a Social Web with Biases


The Web Observatory records behaviour and likes of people on the Web. Such observations may result in intelligent systems that provide recommendations of what to choose or what to do. Web observations also record the biases of people favouring one type of product over another, but also exhibiting biases that (dis) favour people because of their gender, ethinicity or other factors. Intelligent systems that build on such biased data lead to discrimination e.g. affecting job hires, insurance tariffs or credit ratings.

The prohibition of discrimination is a fundamental legal principle protected both at the European level (e.g. Charter of Fundamential Rights) and at the national level (e.g. UK Equality Act of 2010). European Data protection laws are suspicious of automated decision making processes, data maximisation practices, and processing of sensitive data such as sex and ethinicity partly becuase they can amount to discriminations. The simple methods pursued nowadays that ignore sensitive data attributes are not appropriate solutions, because the problem persists, if other data attributes correlate with the sensitive data, e.g. if ethnicity correlates with ZIP codes.


Professor Steffan Staab, Faculty of Physical and Applied Sciences

Dr Sophie Stalla-Bourdillon, Faculty of Business, Law and Art

Dr Laura Carmichael, Faculty of Business, Law and Art


Final report.


Privacy Settings