A new study published on arXiv reveals that dense retrieval systems exhibit bias based on identity signals present in user queries. Researchers found that these systems tend to retrieve documents aligning with the query's political lean and perform worse for queries written in African American Language (AAL) compared to White Mainstream English (WME). The study used controlled synthetic queries and naturalistic queries across political news and consumer-health domains, employing five dense retrievers and a sparse baseline. The findings suggest that these retrieval biases could exacerbate polarization and health disparities. AI
IMPACT Potential to exacerbate polarization and health disparities for marginalized language groups.
RANK_REASON The cluster contains a research paper detailing findings on AI retrieval system bias. [lever_c_demoted from research: ic=1 ai=1.0]
Read on arXiv cs.IR (Information Retrieval) →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
