Scholarlink

<p>As a critical topic in social network analysis, influence maximization (IM) aims to select a subset of nodes from a network to disseminate targeted information as widely as possible. However, traditional IM approaches primarily focus on static networks that cannot depict the evolution of the real-world temporal networks and its corresponding impact on the seed selection. Moreover, existing methods that consider temporal factors frequently struggle with balancing the seed quality and computational efficiency. To fill these gaps, this paper introduces a novel <b>s</b>tructural <b>e</b>volutionary <b>a</b>gent-driven <b>i</b>ncremental <b>u</b>pdating framework based on deep learning, called SEAIU, for the dynamic influence maximization problem. The core of SEAIU lies in utilizing proxies to track topological evolution changes to reduce redundant computations and incrementally update node features to ensure the quality of the seed set as the network evolves. The proposed structural evolution agent strategy selects key nodes that play an important role in the evolutionary process as agents, rather than all nodes in the network. Then proxy nodes use deep learning to recompute embeddings and incrementally update the embeddings of other nodes. Finally, SEAIU uses updated embeddings to select the optimal seeds in the process of network evolution. Extensive experimental results show that this framework can track optimal influential seeds across various types of networks and evolutionary patterns, while significantly reducing redundant computations by an average of 15% in rapidly evolving networks and by 70% in stable networks.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Structural evolutionary agent meets deep learning: a novel incremental updating framework for dynamic influence maximization in temporal social networks

  • 期刊论文

  • 发表日期：

    2026年3月30日

查看全文
  • Jianxin Tang,
  • Jitao Qu,
  • Chenshuo Li,
  • Jian Shi,
  • Chunxia Wang

摘要

As a critical topic in social network analysis, influence maximization (IM) aims to select a subset of nodes from a network to disseminate targeted information as widely as possible. However, traditional IM approaches primarily focus on static networks that cannot depict the evolution of the real-world temporal networks and its corresponding impact on the seed selection. Moreover, existing methods that consider temporal factors frequently struggle with balancing the seed quality and computational efficiency. To fill these gaps, this paper introduces a novel structural evolutionary agent-driven incremental updating framework based on deep learning, called SEAIU, for the dynamic influence maximization problem. The core of SEAIU lies in utilizing proxies to track topological evolution changes to reduce redundant computations and incrementally update node features to ensure the quality of the seed set as the network evolves. The proposed structural evolution agent strategy selects key nodes that play an important role in the evolutionary process as agents, rather than all nodes in the network. Then proxy nodes use deep learning to recompute embeddings and incrementally update the embeddings of other nodes. Finally, SEAIU uses updated embeddings to select the optimal seeds in the process of network evolution. Extensive experimental results show that this framework can track optimal influential seeds across various types of networks and evolutionary patterns, while significantly reducing redundant computations by an average of 15% in rapidly evolving networks and by 70% in stable networks.