An Overlooked Risk: Investigating Name-Based Bias in TF-IDF-Based AI Spam Filters - A Literature Review

Abstract

As artificial intelligence becomes increasingly integrated into digital communication, ensuring that these systems operate fairly has become increasingly important. This literature review examines whether existing research suggests that TF-IDF-based email spam filters may exhibit bias against culturally diverse names by synthesizing studies on spam filtering, machine learning fairness, natural language processing, and name-based bias. The literature shows that machine learning systems can inherit bias from training data and that personal names may function as demographic signals, yet little direct research has examined whether these findings extend to TF-IDF-based spam filters. Rather than concluding that such bias exists, this review identifies an important gap in the literature and argues that current evidence provides a strong theoretical basis for future empirical research evaluating fairness in traditional text classification systems.