This system integrates product tagging and personalized recommendations using large language models, user behavior data, and multimodal learning. It includes two modules: BETag, which generates behavior-aligned tags via LLM fine-tuned on user history, and MTSTRec, which integrates multimodal data through a time-aligned token and Transformer to model user preferences. The system enhances tagging and recommendation accuracy across e-commerce and content platforms and supports scalable deployment.

This system integrates two innovations: BETag and MTSTRec. BETag combines LLMs with user behavior data to generate behavior-enhanced tags, boosting retrieval and recommendation tasks beyond human-annotated and other automated taggers. MTSTRec introduces a Time-Aligned Shared Token mechanism to effectively fuse asynchronous multimodal data, significantly improving recommendation accuracy and outperforming benchmark models.

This technology applies to e-commerce, media, and multimodal-driven industries. It addresses the challenges in product/item tagging and recommendation accuracy. BETag generates behavior-enhanced tags offline, reducing costs and boosting relevance. MTSTRec integrates multimodal data to deliver precise personalized recommendations, boosting conversion rates and user engagement. The system is easy to integrate with existing e-commerce platforms and offers an effective, scalable solution.