Abstract: Collaborative filtering recommendation method performance decreases, when the target rating data is very sparse. The cross domain recommendation method can solve the problem of data sparsity to a certain extent, but for heterogeneous data in different domains, it may lead to negative transfer if no feature mapping processing is performed. Adopting a single transfer model, will cause potential information loss. Therefore, a domain adaptive approach is proposed to apply to cross domain recommendation. The concrete includes：firstly, GFK feature mapping is used to increase the consistency of shared information and reduce the loss of potential information. In order to improve the accuracy of predictions, joint user focus and item focus are used to predict missing rating. Experimental results on open source dataset demonstrate that the proposed model can improve the accuracy of recommendation.

Key words: transfer learning, recommendation technology, domain adaptation, data sparsity, feature mapping

摘要： 协同过滤在目标评分数据非常稀疏时，其推荐效果往往会下降。跨领域推荐方法在一定程度上可以解决数据稀疏性的问题。对于不同领域异构的数据，如果不进行特征映射处理，则可能会导致负迁移；采用单一的迁移模式，则会造成潜在信息缺失。因此，提出一种领域自适应的方法，以应用于跨领域推荐。具体包括：利用GFK特征映射后，以增加共享信息的一致性和减少潜在信息的缺失；采用联合用户侧重和项目侧重多元迁移模式来预测缺失评分的目标域矩阵，以提升预测评分的准确性。经开源数据集上的实验测试，证实了该模型可提高推荐的精准度。

关键词: 迁移学习, 推荐方法, 域自适应, 数据稀疏, 特征映射