Analysis and Research on the Information Narrowing in Personalized Recommendation Systems

In the information age, personalized recommendation systems play a crucial role in addressing the problem of information overload. As deep learning technology has been applied in recommendation systems, the issue of homogeneous content recommendations has become increasingly severe, leading to the phenomenon of information narrowing. In particular, the filter bubbles generated by the computation and decision-making of deep learning algorithms are a typical manifestation of this type of problem. The emergence of filter bubbles limits the diversity of information accessible to users and may influence the quality of their final decisions. Existing studies reveal that introducing models with richer entity representations, like knowledge graph technology, can effectively address the lack of information and sparsity in recommendation systems, while mitigating filter bubble issues. Thus, this paper proposes a multimodal knowledge graph convolutional network model, inspired by knowledge graph convolutional networks. By leveraging entity neighborhood aggregation and introducing the concept of multimodality, using movie recommendations as an example, it enhances the entity representation in recommendations, thereby alleviating the information narrowing problem. The results indicate that the optimized model performs well in terms of recommendation accuracy and also increases the diversity of recommended content, thus providing users with a broader information perspective.