Personalized Recommendation Algorithm in Complex Networks via Multi-Granularity Community Features and Fast Parallel Matrix Factorization

针对复杂网络环境下个性化推荐系统面临的数据稀疏性、计算效率及用户偏好动态性等挑战，本文提出了一种融合多粒度社团特征与快速并行矩阵分解的推荐算法。该算法首先构建基于用户相似度的复杂网络，并引入改进的社团检测算法实现多粒度社团结构划分，生成粗细粒度不同的用户群体；随后，设计了一种基于用户活跃度的自适应层次选择机制，旨在根据用户行为特征动态选择合适的社团层次进行推荐，以缓解数据稀疏性问题并提升推荐精度。在此基础上，本文提出一种基于多层用户相似性网络的快速并行梯度下降矩阵分解模型，有效捕捉用户在不同粒度上的偏好特征，并结合用户全局偏好信息与局部社团特征。多个真实数据集上的实验结果表明，相较于传统推荐算法，融合多粒度社团特征与快速并行矩阵分解算法在准确性、精确率和召回率等指标上均有不同程度的提升，验证了该算法在提高推荐质量和可扩展性方面的有效性。To address the challenges of data sparsity, computational efficiency, and dynamic user preferences in personalized recommendation systems within complex networks, this paper proposes a recommendation algorithm that integrates multi-granularity community features with fast parallel matrix factorization. First, a complex network based on user similarity is constructed, and an improved community detection algorithm is introduced to achieve multi-granularity community structure partitioning, generating user groups at different levels of granularity. Then, an adaptive hierarchical selection mechanism based on user activity is designed to dynamically select the appropriate community level for recommendation according to user behavior characteristics, thereby alleviating data sparsity and improving recommendation accuracy. On this basis, we propose a fast parallel gradient descent matrix factorization model based on multi-layer user similarity networks, effectively capturing user preference features at different granularities while incorporating both global user preference information and local community characteristics. Experimental results on multiple real-world datasets demonstrate that, compared with traditional recommendation algorithms, the proposed approach achieves significant improvements in accuracy, precision, and recall by integrating multi-granularity community features with fast parallel matrix factorization, verifying its effectiveness in enhancing recommendation quality and scalability.