%0 Journal Article %T 融合网络社团特征与矩阵分解的推荐算法
Recommendation Algorithm Based on Fast Parallel Stochastic Gradient Descent Matrix Factorization and User Similarity Network Community Features %A 艾均 %A 李贵平 %A 苏湛 %A 苏杭 %J Modeling and Simulation %P 699-715 %@ 2324-870X %D 2025 %I Hans Publishing %R 10.12677/mos.2025.144322 %X 为解决推荐系统在处理超大规模数据集时计算复杂度高、资源消耗大，以及推荐性能与可扩展性之间的矛盾，本文设计了一种融合用户相似性网络社团特征的快速并行梯度下降矩阵分解算法。首先，通过计算用户相似性，基于高相似度用户关系构建相似性网络，并采用模块度优化算法将用户划分为若干社团。同时，设计了一种用户补偿机制，在社团划分中引入指定数量的活跃用户，降低基于用户相似性社团划分的用户群内部数据的过度相似性。随后，针对每个用户群，采用并行梯度下降矩阵分解算法学习用户与物品的隐因子，以预测用户对未知物品的评分。该模型通过社团分组和并行计算，不仅减少了计算复杂度和资源消耗，还有效提升了推荐系统的准确性和算法学习效率。与现有领先算法相比，该模型在多个数据集上有效提高了推荐精度和召回率，降低了均方根误差和平均绝对误差。
To address the challenges of high computational complexity and resource consumption in recom- mendation systems when handling ultra-large datasets, this study proposes a fast parallel gradient descent matrix factorization algorithm that integrates user similarity network community features. First, by calculating user similarities, a similarity network is constructed based on highly similar user relationships, and a modularity optimization algorithm is employed to partition users into several communities. Additionally, a user compensation mechanism is designed to introduce a specified number of active users into the community division, thereby reducing the excessive similarity of data within user groups based on similarity community partitioning. Subsequently, for each user group, a parallel gradient descent matrix factorization algorithm is utilized to learn the latent factors of users and items, enabling the prediction of user ratings for unknown items. This model reduces computational complexity and resource consumption through community grouping and parallel computation, while effectively enhancing the accuracy of the recommendation system and the efficiency of algorithm learning. Compared to existing leading algorithms, this model significantly improves recommendation precision and recall across multiple datasets, while also lowering root mean square error and mean absolute error. %K 用户相似性网络社团发现， %K 梯度下降矩阵分解， %K 可扩展性， %K 数据稀疏， %K 推荐算法， %K 社团检测算法
User Similarity Network Community Discovery %K Gradient Descent Matrix Factorization %K Scalability %K Data Sparsity %K Recommendation Algorithm %K Community Detection Algorithm %U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=112472