Collaborative Recommendation of Innovation, Entrepreneurship, and Professional Courses Using a Multimodal ViT Transformer Fusion Model

In the context of the deep integration of innovation and entrepreneurship education with professional education, existing course recommendation systems still suffer from single data sources, insufficient feature mining, and weak collaboration among heterogeneous courses. These limitations hinder the effective alignment between curriculum resources and students’ comprehensive development needs. To address these issues, this paper proposes a collaborative recommendation method based on a multimodal ViT–Transformer fusion model. A multimodal dataset integrating structured course and academic data, unstructured textual information, and visual teaching resources is first constructed. Visual features are extracted using the ViT model, textual semantic features are mined through Transformer encoders, and structured data are processed by embedding layers to achieve deep representation of multidimensional features. An attention-based feature interaction module is further designed to strengthen the correlation mining between innovation and entrepreneurship courses and professional courses, thereby improving collaborative matching across course types. Experiments based on real teaching scenarios verify the effectiveness of the proposed model in recommendation accuracy, recall, F1-score, and synergy indicators. The method also provides a useful reference for multimodal information fusion and heterogeneous information propagation in intelligent engineering education systems.