【摘要】 推荐算法能够有效提升信息传递效率。目前，基于图卷积网络(GCN)的协同过滤推荐模型通常未考虑每个节点所扮演角色具有不同的重要性和权重；此外，在模型损失优化过程中容易出现样本不平衡，负样本远多于正样本，导致模型存在一定程度上的过拟合问题，限制了推荐性能。为了更好地提高推荐算法的推荐性能，提出一种基于自适应权重和随机负采样的图卷积网络推荐模型(ANS-GCN)。该模型可以在模块中计算节点权重，以捕获不同节点的重要性权重，是一种方便的即插即用方法；在损失优化模块融合了随机负采样策略作为辅助损失，缓解训练过程中样本类别不平衡问题，降低模型过拟合程度，提高泛化能力。在3个公开数据集上进行对比试验，结果表明，本文模型在Recall@20和NDCG@20两个评价指标上均优于基线模型。更多还原

【Abstract】 Recommendation algorithms can significantly enhance the efficiency of information dissemination.However, current GCN-based collaborative filtering recommendation models often fail to account for the varying importance and weights of roles played by each node.Additionally, sample imbalance, characterized by a far greater number of negative samples than positive ones, frequently occurs during model loss optimization.This imbalance leads to overfitting and restricts the recommendation performance.To tackle these challenges, this paper introduces an Adaptive Negative Sampling Graph Convolutional Network(ANS-GCN) recommendation model that incorporates adaptive weights and random negative sampling.The model calculates node weights within its modules to capture the varying importance of different nodes, offering a convenient plug-and-play solution.Furthermore, it integrates a random negative sampling strategy into the loss optimization module as auxiliary loss, which alleviates the problem of sample class imbalance during training, reduces overfitting, and enhances generalization performance.Comparative experiments on three public datasets demonstrate that the proposed ANS-GCN model outperforms baseline models in terms of both Recall@20 and NDCG@20 metrics.更多还原