改进的知识图注意力推荐模型
An improved recommendation model of knowledge graph attention
石瑞雪 1温秀梅 2严鑫瑜 1陈威1
作者信息
- 1. 河北建筑工程学院,河北 张家口 075000
- 2. 河北建筑工程学院,河北 张家口 075000;张家口市大数据技术创新中心,河北 张家口 075000
- 折叠
摘要
针对知识图注意力网络(Knowledge Graph Attention Network,KGAT)推荐模型在整个知识图谱上传播信息时,容易丢失大量特征信息的问题,提出一种改进的知识图注意力网络模型,通过将注意力机制替换为双向注意力机制,从而提高了推荐的准确率.最后,在两个公开数据集Amazon-Book和Last-FM上进行了对比实验,实验结果表明:改进的模型在 Re-call和NDCG 两项评价指标上都得到了提高,在 Amazon-Book 上,分别提高了 1.81%和1.68%;在Last-FM上,分别提高了 1.26%和 1.35%,有效地改善了推荐效果.
Abstract
Because of the problem that the knowledge graph attention network(Knowledge Graph At-tention Network,KGAT)recommendation model spreads information on the whole knowledge graph and is easy to lose a large amount of feature information,an improved knowledge graph at-tention network model is proposed,which replaces the attention mechanism by a two-way atten-tion mechanism to improve the accuracy of recommendation.Finally,comparative experiments were conducted on two public datasets,Amazon-Book and Last-FM.The experimental results showed that the improved model was improved in the evaluation indexes of Recall and NDCG,by 1.81%and 1.68%,respectively,and 1.26%and 1.35%on Last-FM,respectively,effectively im-proving the recommendation effect.
关键词
知识图注意力网络/注意力机制/推荐系统/知识图谱Key words
Knowledge graph attention network/Attention mechanism/Recommendation system/Knowledge graph引用本文复制引用
出版年
2024
NETL
NSTL
万方数据