【摘要】 针对基于映射的跨域推荐模型未充分关注源域中数据稀疏的用户，导致用户偏好的迁移效率降低的问题，提出了一种基于邻居交互增强和多头注意力机制的跨域推荐模型。首先，利用邻居用户的交互项目来增强源域中数据稀疏用户的交互序列，以捕获更丰富的用户行为信息。然后，采用多头注意力机制从交互序列中提取用户可迁移的偏好特征，以全面捕捉用户兴趣的多个方面。最后，将提取的用户特征输入元网络生成个性化映射函数，并根据源域迁移的用户嵌入来实现目标域的个性化推荐。在亚马逊和豆瓣数据集上进行实验，结果表明所提出的模型相较于最优的基线模型，平均绝对误差指标最高提升了6.54%,均方根误差指标最高提升了3.73%。有效地提高了目标域的推荐性能，能够在电子商务等领域为用户提供更准确的项目推荐。更多还原

【Abstract】 To address the problem that the mapping-based cross-domain recommendation model does not pay enough attention to the users with sparse data in the source domain, resulting in a decrease in the transfer efficiency of user preferences, we propose a cross-domain recommendation model based on neighbor interaction enhancement and multi-head attention mechanism is proposed.Firstly, the interaction items of neighbor users are used to enhance the interaction sequences of data-sparse users in the source domain, capturing richer user behavior information.Then, a multi-head attention mechanism is adopted to extract transferable user preference features from the interaction sequence, comprehensively capturing multiple aspects of user interests.Finally, the extracted user features are input into the meta-network to generate a personalized mapping function, enabling personalized recommendations in the target domain based on the transferred embeddings of users from the source domain.The experimental results on Amazon and Douban datasets demonstrate that the proposed model outperforms the best baseline model, with a maximum improvement of 6.54% in the mean absolute error metric and 3.73% in the root mean square error metric.This effectively enhances the recommendation performance in the target domain, enabling more accurate item recommendations for users in areas like e-commerce.更多还原