A knowledge graph-integrated recommendation method for college student career planning

Abstract Facing the complex and diverse interests of college students, current career planning recommendation methods often suffer from low accuracy. To tackle this, we propose a novel knowledge graph-integrated recommendation model. Our approach begins by constructing a comprehensive career planning knowledge graph. We then project sparse user and item vectors into a dense latent space. Leveraging triplet information from the graph, we form project neighborhoods and employ a Graph Convolutional Network to adaptively aggregate neighbor vectors, deriving enriched representations for students and careers. A key innovation is the fusion of these graph-based representations with those from a generalized matrix factorization component. The concatenated vectors are fed into a fully connected layer to output the final recommendation score. Extensive comparative experiments on three real-world career planning datasets demonstrate the superiority of our model, showing significant improvements in recommendation accuracy of at least 2.04%, 2.26%, and 1.34%, respectively, over state-of-the-art methods.