Kaili Liao

With the explosive growth of short video traffic and the increasing demand for low-latency content delivery, efficient edge caching strategies have become critical for mobile networks. However, the highly dynamic and personalized characteristics of short video services present substantial challenges for traditional caching approaches, which often depend exclusively on static popularity metrics. This paper proposes DECC (Dynamic Edge-caching through Content Popularity and Crowd Prediction), a novel caching framework that jointly models content popularity and user access behavior to optimize caching decisions at edge nodes. DECC integrates a hybrid deep learning architecture comprising 1D Convolutional Neural Networks (Conv1D), Long Short-Term Memory (LSTM) networks, and Gated Recurrent Units (GRU) to capture the temporal dynamics of both video requests and user activity. A fusion mechanism is introduced to generate cache priority scores based on dual-path predictions, enabling more accurate and adaptive content placement. Experimental evaluations conducted on real-world datasets demonstrate that DECC consistently surpasses baseline methods in cache hit rate, access latency reduction, and overall resource utilization efficiency. These results highlight the potential of DECC as a scalable and intelligent caching solution for next-generation short video edge services.

Information popularity prediction is a critical problem in social network analysis. With the increasing prevalence of social platforms, accurate prediction of the diffusion process has become increasingly important. Existing methods mainly rely on graph neural networks to model structural relationships, but they are often insufficient in capturing the complex interplay between temporal evolution and local cascade structures, especially in real-world scenarios involving sparse or rapidly changing cascades. To address this issue, we propose the Cascading Dynamic attention-calibrated Graph Convolutional Network, named CasDacGCN. It enhances prediction performance through spatiotemporal feature fusion and adaptive representation learning. The model integrates snapshot-level local encoding, global temporal modeling, cross-attention mechanisms, and a hypernetwork-based sample-wise calibration strategy, enabling flexible modeling of multi-scale diffusion patterns. Results from experiments demonstrate that the proposed model consistently surpasses existing approaches on two real-world datasets, validating its effectiveness in popularity prediction tasks.

In recent years, the deployment of edge servers has attracted significant research interest, with a focus on maximizing their utilization under resource constraint to improve overall efficiency. However, most existing studies concentrate on initial deployment strategies, paying limited attention to approaches involving incremental expansion. As user demands continue to escalate, many edge systems are facing overload situations that hinder their ability to meet performance requirements. To tackle these challenges, this paper introduces an auxiliary edge-server deployment strategy designed to achieve load balancing across edge systems and alleviate local server overloads. The problem is herein referred to as the Auxiliary Edge Server Deployment (A–ESD) problem, and the aim is to determine the optimal deployment scheme for auxiliary edge servers. A–ESD is modeled as a multi-objective optimization problem subject to global constraints and is demonstrated to be NP-hard. An enhanced genetic algorithm called LBA–GA is proposed to efficiently solve the A–ESD problem. The algorithm is designed to maximize overall load balance while minimizing total system delay. Extensive experiments conducted on real-world datasets demonstrate that LBA–GA outperforms existing methods, delivering superior load balancing, reduced latency, and higher cost-effectiveness.