Дослідження трансформерних моделей машинного навчання у секвенційних рекомендаційних системах

Abstract

This work is devoted to transformer-based models in sequential recommender systems. The problem of modeling user interaction sequences is considered. Existing approaches, including recurrent, convolutional and transformer models, are analyzed. A comparative evaluation is performed using Hit Ratio and Normalized Discounted Cumulative Gain. The results show that transformer models achieve higher recommendation accuracy than recurrent and convolutional approaches. Autoregressive transformers provide a balanced trade-off between recommendation quality and computational complexity.