Identifying Inequality in Recommendations: A Framework and Experimental Study
Abstract
Recommender systems have become increasingly prevalent online within the last two decades and are used extensively on social media and networking platforms. We look at one such platform, LinkedIn, which serves the purpose of matching job candidates to open jobs and connecting recruiters to job seekers. More specifically, we look into the approaches used to design recommendation systems and analyze how LinkedIn’s search and recommendation algorithms work. However, an issue that arises from the use of recommender systems for job searching is that certain jobs may be shown more often to candidates based on their descriptions, possibly resulting in bias when it comes to networking and potential hiring chances. Because of the impact this could have on both companies and candidates, it is important to ensure that recommendations resulting from these systems are not biased or unfair. Through this thesis, we propose a framework for the systematic study of inequality that manifests in recommendation results. In particular, we focus on inequality in job recommendations, where we quantify this inequality in terms of the Gini coefficient. In our preliminary experimental studies, we find that some factors may have a greater influence on the recommendations made by the algorithm than others. Notably, we observe that many job recommendations show high degrees of concentration with respect to companies, and we hope to build upon this work to explore inequality in terms of other factors as well.
Subject
inequalityrecommender systems
framework
jobs
search queries
recommendation
Gini coefficient
company distribution
Citation
Kohli, Diva (2022). Identifying Inequality in Recommendations: A Framework and Experimental Study. Undergraduate Research Scholars Program. Available electronically from https : / /hdl .handle .net /1969 .1 /196599.