Multimodal Hybrid Healthcare Recommendation System Based on ERT-MOE and Large Language Model enhancement

With the explosion of information and the development of information technology, advanced architectures that can effectively integrate and process diverse data sources have attracted widespread attention and demand in the industry. This study proposes a recommendation model for healthcare with a hybrid architecture, which integrates the most advanced large language model and multimodal data, and introduces an entity recognition transformer (ERT) to extract entities and relations in patient descriptions, and introduces a MOE model to enhance the classification performance of the neural matrix factorization model. This study uses a small dataset to test the recommendation of healthy food based on user symptom descriptions. The experimental results show that the hybrid model outperforms the baseline model of its components in all indicators of personalized recommendation effects. The research results provide an effective way to develop customized recommendation systems and promote the constructive contribution of recommendation technology in practical fields such as guidance and self-examination.