Gender Stereotype Reinforcement: Measuring the Gender Bias Conveyed by Ranking Algorithms

TL;DR

This paper introduces GSR, a novel measure to quantify gender stereotype reinforcement in search engine rankings, validating it on synthetic and real data, and analyzing how different algorithms and debiasing methods influence gender bias.

Contribution

The paper presents GSR, the first IR-specific measure for quantifying gender stereotype reinforcement in ranking algorithms, and evaluates bias across various models and debiasing techniques.

Findings

GSR effectively measures gender bias in search rankings.

Ranking algorithms inherit biases from word embeddings.

Debiasing methods reduce GSR but may impact performance.

Abstract

Search Engines (SE) have been shown to perpetuate well-known gender stereotypes identified in psychology literature and to influence users accordingly. Similar biases were found encoded in Word Embeddings (WEs) learned from large online corpora. In this context, we propose the Gender Stereotype Reinforcement (GSR) measure, which quantifies the tendency of a SE to support gender stereotypes, leveraging gender-related information encoded in WEs. Through the critical lens of construct validity, we validate the proposed measure on synthetic and real collections. Subsequently, we use GSR to compare widely-used Information Retrieval ranking algorithms, including lexical, semantic, and neural models. We check if and how ranking algorithms based on WEs inherit the biases of the underlying embeddings. We also consider the most common debiasing approaches for WEs proposed in the literature and test their impact in terms of GSR and common performance measures. To the best of our knowledge, GSR is the first specifically tailored measure for IR, capable of quantifying representational harms.