The enormous amount of network equipment and users implies a tremendous growth of Internet traffic for multimedia services. To mitigate the traffic pressure, architectures with in-network storage are proposed to cache popular content at nodes in close proximity to users to shorten the backhaul links. Meanwhile, the reduction of transmission distance also contributes to the energy saving. However, due to limited storage, only a fraction of the content can be cached, while caching the most popular content is cost-effective. Correspondingly, it becomes essential to devise an effective popularity prediction method. In this regard, existing efforts adopt dynamic graph neural network (DGNN) models, but it remains challenging to tackle sparse datasets. In this paper, we first propose a reformative temporal graph network, which is named STGN, that utilizes extra semantic messages to enhance the temporal and structural learning of a DGNN model, since the consideration of semantics can help establish implicit paths within the sparse interaction graph and hence improve the prediction performance. Furthermore, we propose a user-specific attention mechanism to fine-grainedly aggregate various semantics. Finally, extensive simulations verify the superiority of our STGN models and demonstrate their high potential in energy-saving.

翻译：网络设备及用户的巨大规模意味着多媒体服务流量的急剧增长。为缓解流量压力，具备网络内存储能力的架构被提出，通过在靠近用户的节点处缓存热门内容以缩短回程链路；同时，传输距离的缩减也有助于节能。然而，受限于有限的存储空间，仅能缓存部分内容，而缓存最热门的内容更具成本效益。因此，设计有效的流行度预测方法至关重要。现有相关研究采用动态图神经网络模型，但在处理稀疏数据集时仍面临挑战。本文首先提出一种改良的时序图网络——STGN，它利用额外的语义信息增强动态图神经网络的时序与结构学习能力——因为语义的引入有助于在稀疏交互图中建立隐式路径，从而提升预测性能。此外，我们设计了一种用户特定注意力机制，实现多类语义的细粒度聚合。最终，大量仿真验证了STGN模型的优越性，并表明其在节能方面具有巨大潜力。