Classifying Tenant Legal Inquiries: A Comparative Study of Traditional and Deep Learning Approaches

Legal aid organizations face increasing demand for tenant protection services amid limited resources. Accurate classification of tenant inquiries enables efficient case routing and volunteer attorney matching. This study compares traditional machine learning methods (Naive Bayes, Support Vector Machines) with deep learning approaches (BERT fine-tuning) for classifying tenant legal inquiries across four categories: illegal eviction, housing repairs, security deposit disputes, and rent disagreements. Experiments on 450 de-identified tenant assistance requests reveal that sample size critically impacts method selection. Traditional approaches demonstrate robust performance with fewer than 150 samples, while BERT achieves superior accuracy (F1-score 0.89 vs 0.81) with datasets exceeding 300 samples. Mixed-issue cases involving multiple complaint types pose consistent challenges across all methods. Results inform practical deployment strategies for legal aid intake workflows.