Replication Study: Classifying User Requirements from Online Feedback in Small Dataset Environments using Deep Learning

This research investigates automated classification of user requirements from app reviews, comparing BERT-based models with traditional machine learning and deep learning approaches. The study introduces a novel application of GPT-4 for zero-shot classification using carefully engineered prompts and deterministic settings. Through comprehensive analysis of 9,650+ reviews, the work evaluates different model architectures while addressing practical implementation challenges in dependency management and dataset quality. The findings demonstrate the effectiveness of transformer-based models for requirements classification and highlight key considerations for real-world deployment. The research provides both methodological insights into prompt design and statistical analysis, along with practical frameworks that can be adapted for industry applications, while identifying promising directions for future work in multi-label classification and domain adaptation.