Enhancing Movie Recommendations in Fully Automated Vehicles: A Multi-Interest Approach With Transformer Models

While many existing movie recommendation systems have been integrated to personalize entertainment in FAVs, they face challenges in addressing the diverse and dynamic preferences of multiple passengers. To tackle these issues, this article introduces the multigate mixture of experts for multiinterest model (MEMI) specifically designed for FAVs. The proposed model employs a Transformer-based multi-interest extractor within a multigate mixture of experts (MMoE) structure to capture a range of person interests while managing network complexity. Additionally, a novel peak interest alignment (PIA) loss function is introduced to improve consistency between the training and inference phases, ensuring more accurate recommendations. Experimental evaluations using the Movielens dataset demonstrate that the proposed model significantly outperforms existing systems, providing more personalized and effective movie recommendations.