摘要
针对传统推荐模型面临的数据稀疏性问题,提出一种基于结合注意力机制的门控循环单元的融合语义和知识特征的推荐模型。基于知识图谱,使用连续词袋模型捕获项目实体对应的语义特征,依据“偏好扩散”思想进行知识特征的学习,将不同层面特征进行融合后,使用结合注意力机制的门控循环单元挖掘用户潜在兴趣偏好。基于MovieLens数据集的对比实验结果表明,所提模型能够有效提升推荐效果并缓解数据稀疏性问题,通过消融实验验证了该模型各个组件的有效性。
To address the problem of data sparsity faced by traditional recommendation models,a recommendation model combining semantic feature and knowledge feature based on gated recurrent unit with attention mechanism was proposed.Based on knowledge graph,semantic feature and knowledge feature of the item entities were obtained with the continuous bag-of-words model and the idea of preference propagation.After integrating different features,the gated recurrent unit with the attention mechanism was used to capture potential interest preferences of users.Results of the comparative experiments on the MovieLens indicate that the proposed model can improve the performance of recommendation results and effectively alleviate influences of the data sparsity.The effectiveness of each component of the model is verified by ablation experiments.
作者
郑光
朱越
时雷
马新明
席磊
ZHENG Guang;ZHU Yue;SHI Lei;MA Xin-ming;XI Lei(College of Information and Management Science,Henan Agricultural University,Zhengzhou 450046,China;Henan Engineering Laboratory for Farmland Environmental Monitoring and Control Technology,Henan Agricultural University,Zhengzhou 450002,China)
出处
《计算机工程与设计》
北大核心
2023年第8期2506-2515,共10页
Computer Engineering and Design
基金
国家重点研发计划课题基金项目(2016YFD0300609)。
关键词
推荐模型
知识图谱
特征融合
门控循环单元
注意力机制
语义特征
连续词袋
recommendation model
knowledge graph
feature fusion
gated recurrent unit
attention mechanism
sematic feature
continuous bag-of-words
