面向稀疏数据场景的生成对抗网络推荐算法

基于生成对抗网络的协同过滤算法(CFGAN)是生成对抗网络在个性化推荐领域上的重大突破，但CFGAN存在缺乏对用户可能交互物品的关注以及面对稀疏数据场景特征提取能力较弱的问题。另外，结合负采样技术提出的优化CFGAN方案存在无法结合用户属性抽取负样本的缺陷。为此，本文提出一种改进的CFGAN模型，通过引入增强的置换注意力机制强化面向稀疏数据集的特征聚焦能力，同时考虑用户可能交互物品对推荐结果的影响；此外，采用协同用户社交网络从用户反馈中提取的语义好友特征嵌入CFGAN，以实现负样本的个性化抽取，进一步提升模型面向稀疏数据场景的推荐效果。

A generative adversarial network recommendation algorithm for sparse data scenarios

The collaborative filtering framework based on generative adversarial network (CFGAN) is a breakthrough in the field of generative adversarial networks for personalized recommendation, but CFGAN suffers from a lack of focus on the possible interaction items of users and weak feature extraction ability in the sparse data scenarios. In addition, the proposed optimized CFGAN scheme combined with negative sampling has the drawback of not being able to extract negative samples by combining user attributes. Therefore, an improved CFGAN model is proposed to enhance the feature focus capability for sparse datasets by introducing an enhanced shuffle attention mechanism, while taking into account the influence of possible user interaction items on the recommendation results. Furthermore, the semantic friend features extracted from user feedback by collaborative user social networks are embedded into CFGAN to achieve personalized extraction of negative samples, further enhancing the model''s recommendation effectiveness for sparse data scenarios.