Popularity Bias as Ethical and Technical Issue in Recommendation: A Survey

Abstract

Recommender Systems have become omnipresent in our everyday life, helping us making decisions and navigating in the digital world full of information. However, only recently researchers have started discovering undesired and harmful effects of automated recommendation and began questioning how fair and ethical these systems are, while influencing our day-to-day decision making, shaping our online behaviour and tastes. In the latest research works, various biases and phenomena like filter bubbles and echo chambers have been uncovered among the resulting effects of recommender systems and rigorous work has started on solving these issues. In this narrative survey, we investigate the emergence and progression of research on one of the potential types of biases in recommender systems, ie Popularity Bias. Many recommender algorithms have been shown to favor already popular items, hence giving them even more exposure, which can harm fairness and diversity on the platforms using such systems. Such a problem becomes even more com-plicated if the object of recommendation is not just products and content, but people, their work and services. This survey describes the progress in this field of study, highlighting the advancements and identifying the gaps in the research, where additional effort and attention is necessary to minimize the harmful effect and make sure that such systems are build in a fair and ethical way.