The research on microscopic diffusion prediction is of great significance for understanding the propagation of information in social networks. To improve the accuracy of information diffusion predictions, an information microscopic diffusion prediction model integrating cascaded frequency domain features is proposed. First, a social graph and an information diffusion hypergraph are constructed separately based on user friendship relationships and historical cascades. Graph convolutional neural networks are utilized to capture user representations in social relationships and forwarding behaviors. Next, Fourier Transform is applied to map the time-domain cascade features to the frequency domain, effectively capturing both short-term fluctuations and long-term trends in the cascade through high-frequency and low-frequency components. Finally, an attention fusion layer is designed to generate a more expressive user representation, effectively addressing the issues of feature redundancy and information loss. Thus, the performance of the proposed model is further optimized. Experiments on four public datasets show the proposed model improves Hits@K and mAP@K, demonstrating the effectiveness of itself.

The interaction between microorganisms and drugs significantly impacts human health. In existing association prediction methods, the internal node information of heterogeneous graphs is not adequately modeled and the importance of different meta-path instances is overlooked. Hence, a biological topology-semantic enhanced heterogeneous graph representation learning for drug-microbe interactions(HGRL) method is proposed. High-order mixed neighborhood network embedding representations are extracted to infer microorganism-drug associations. Microorganism-drug similarity and association data are integrated to construct a weighted bidirectional heterogeneous network and a multi-view meta-path aware network. The transformer-gated graph network is combined with Bayesian Gaussian mixture weighted contrastive learning to extract topological semantics and embedding features of complex biological networks. Prediction based on adversarial negative sampling demonstrates that HGRL outperforms existing methods in microorganism-drug association prediction and is a reliable tool for inferring microorganisms associated with candidate drugs.