Scholarlink

<p>Community detection is a critical challenge in analyzing complex networks, with traditional methods often struggling with scalability and accuracy in large and intricate systems. This paper proposes a novel hybrid approach, MGO-DE-LA-BP-CDDL, to address these issues. This method adapts multiple techniques, including Mountain Gazelle Optimizer (MGO), Differential Evolution (DE), Learning Automata (LA), and Deep Backpropagation (Deep BP). The framework utilizes MGO-DE-LA for global optimization while using the BP technique for local refinement, ultimately enhancing community detection accuracy. The method incorporates a self-adaptive mechanism that dynamically balances exploration and exploitation, enabling robust optimization and uncovering hidden network structures. Parallelized and distributed computations on CPUs and GPUs further improve scalability and efficiency. The proposed method employs two objective functions to guide the optimization process and integrates stacked unsupervised learning for feature extraction with supervised BP algorithms for refinement. Experimental evaluations on real-world datasets demonstrate that the proposed method significantly outperforms traditional approaches, achieving higher modularity scores and improved community detection accuracy. These results highlight the proposed method’s potential to advance network science by providing a scalable and efficient solution for analyzing complex systems.</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

A hybrid deep learning and metaheuristic optimization algorithm for accurate and scalable community detection in complex networks

  • 期刊论文

  • 发表日期：

    2026年1月27日

查看全文
  • Solmaz Pourmahmoud,
  • Farhad Soleimanian Gharehchopogh,
  • Mohammad Masdari

摘要

Community detection is a critical challenge in analyzing complex networks, with traditional methods often struggling with scalability and accuracy in large and intricate systems. This paper proposes a novel hybrid approach, MGO-DE-LA-BP-CDDL, to address these issues. This method adapts multiple techniques, including Mountain Gazelle Optimizer (MGO), Differential Evolution (DE), Learning Automata (LA), and Deep Backpropagation (Deep BP). The framework utilizes MGO-DE-LA for global optimization while using the BP technique for local refinement, ultimately enhancing community detection accuracy. The method incorporates a self-adaptive mechanism that dynamically balances exploration and exploitation, enabling robust optimization and uncovering hidden network structures. Parallelized and distributed computations on CPUs and GPUs further improve scalability and efficiency. The proposed method employs two objective functions to guide the optimization process and integrates stacked unsupervised learning for feature extraction with supervised BP algorithms for refinement. Experimental evaluations on real-world datasets demonstrate that the proposed method significantly outperforms traditional approaches, achieving higher modularity scores and improved community detection accuracy. These results highlight the proposed method’s potential to advance network science by providing a scalable and efficient solution for analyzing complex systems.