Scholarlink

The influence maximization problem is a critical task pertaining to social networks, aiming to identifying a set of key individuals who can achieve the maximal information spread through the system. Great attention has been drawn in related studies, and traditional methods often struggle with prohibitive computational efficiency and limited scalability. Therefore, in this paper, a novel approach is developed to solve this problem by leveraging a Prioritized Experience Replay (PER) method within the framework of Deep Reinforcement Learning (DRL), named PER-GDRL. This approach includes a specialized PER method to enhance the learning efficiency and performance of DRL agents. More informative experiences can be prioritized in the training process. Additionally, principles of the PER method are deliberated to optimize the Markov process, making an efficient seed search and selection process. Experimental results demonstrate that PER-GDRL outperforms existing techniques in terms of both performance and computational efficiency. A promising direction is given for applying advanced reinforcement learning techniques to optimization problems related to complex networks.

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

A Genetic Deep Reinforcement Learning Approach with Prioritized Experience Replay Strategy to Solve the Influence Maximization Problem on Networks

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Zhixi Chen,
  • Junge Wu,
  • Shuai Wang,
  • Jing Huang

摘要

The influence maximization problem is a critical task pertaining to social networks, aiming to identifying a set of key individuals who can achieve the maximal information spread through the system. Great attention has been drawn in related studies, and traditional methods often struggle with prohibitive computational efficiency and limited scalability. Therefore, in this paper, a novel approach is developed to solve this problem by leveraging a Prioritized Experience Replay (PER) method within the framework of Deep Reinforcement Learning (DRL), named PER-GDRL. This approach includes a specialized PER method to enhance the learning efficiency and performance of DRL agents. More informative experiences can be prioritized in the training process. Additionally, principles of the PER method are deliberated to optimize the Markov process, making an efficient seed search and selection process. Experimental results demonstrate that PER-GDRL outperforms existing techniques in terms of both performance and computational efficiency. A promising direction is given for applying advanced reinforcement learning techniques to optimization problems related to complex networks.