基于贝叶斯个性化排名和信息传播的TOP-K推荐算法

Top-K Recommendation Algorithm Based on Bayesian Personalized Ranking and Information Dissemination

图神经网络已成为推荐系统领域研究的热点。图神经网络固有的特性使其在训练时存在过平滑的问题。针对这一问题提出了一种基于贝叶斯个性化排名和信息传播的top-k推荐算法。该方法首先使用随机初始化对所有的用户和项目进行嵌入编码,然后将编码输入简化过的图神经网络进行信息传播并训练。每当达到一定的训练次数后将嵌入向量跳过信息传播,使用基于贝叶斯个性化排名的矩阵分解进行训练。最后使用内积计算用户对项目的偏好程度并生成top-k推荐列表。实验结果表明,本算法可以有效缓解图神经网络的过平滑问题,并提高推荐结果的质量。

In recent years, graph neural networks have become a hot research topic in the field of recommendation systems, but because of their inherent characteristics, graph neural networks suffer from the problem of oversmoothing when trained.To address this problem this paper proposes a top-k recommendation algorithm based on Bayesian personalized ranking and information propagation.The method first uses random initialization to embed codes for all users and items, and then feeds the codes into a simplified graph neural network for information propagation and training.Whenever a certain number of training sessions is reached, the embedding vector is skipped and trained using a matrix decomposition based on Bayesian personalized ranking.Finally, the inner product is used to calculate the user’s preference level of items and generate the top-k recommendation list.The experimental results show that this algorithm can effectively alleviate the oversmoothing problem of graph neural networks and improve the quality of recommendation results.