Using Embedding Models to Improve Probabilistic Race Prediction

2026-04-24

AI Summary

This study addresses the significant performance degradation of traditional surname-based race prediction methods, such as Bayesian Improved Surname Geocoding (BISG), when applied to individuals with rare surnames—a limitation stemming from their reliance on census data and the use of uninformative priors for uncovered surnames. To overcome this, the authors propose embedding-enhanced BISG (eBISG), which integrates pretrained name text embeddings with neural networks to model interactions among all components of full names, generating dense vector representations. Leveraging 2020 U.S. Census and Southern state voter registration data, eBISG substantially improves race probability estimation for unseen surname groups, particularly among Asian and Hispanic populations. The variant incorporating full-name embeddings demonstrates the strongest performance, offering a marked advancement over the conventional BISG framework.

Technology Category

• Machine Learning: Semi-Supervised Learning
• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

Estimating racial disparity requires individual-level race data, which are often unavailable due to the sensitivity of collecting such information. To address this problem, many researchers utilize Bayesian Improved Surname Geocoding (BISG), which have critically relied on Census surname data. Unfortunately, these data capture race-surname relationships only for common surnames, omitting approximately 10% of the US population. We show that predictive performance degrades substantially for individuals with such omitted, uncommon surnames because standard BISG implementation relies on a uninformative generic prior in these cases. To address this limitation, we propose embedding-powered BISG (eBISG), which uses pre-trained text embeddings to represent names as dense vectors and trains neural networks on 2020 Census surname and first-name data to estimate race probabilities for names not covered in the Census. We compare five approaches: standard BISG using only surnames, BIFSG incorporating first name probabilities, surname embedding for unlisted names, surname and first name embedding combining both, and a full-name embedding trained on voter file data from Southern states that captures interactions between name components. We show that each successive eBISG approach improves race prediction, with the full-name embedding yielding the largest gains, particularly for Hispanic and Asian voters whose surnames are absent from the Census list.

Problem

• race prediction
• surname data
• racial disparity
• uncommon surnames
• BISG

Innovation

• embedding models
• race prediction
• BISG
• neural networks
• name representation