摘要
在推荐系统中，基于知识图谱的神经网络推荐以图作为输入，可以很好地将节点信息和拓扑结构相结合进行预测。然而现有方法中，很少考虑图结构中存在的对称关系以及信息聚合时梯度消失的问题。本文提出一种双向注意力机制的知识图谱神经网络推荐算法，首先将图神经网络与对称注意力机制相结合，然后采用双向翻译模型对知识图谱中用户-项目信息进行特征的嵌入表示，使得注意力机制在决策权重时考虑的关系更全面。其次，在对节点和邻居信息训练过程中，为避免过拟合问题引入了多通道激活函数。最后，在两个真实数据集上与经典算法进行对比，验证了本文所提出算法的有效性。
关键词： 双向嵌入；注意力机制；知识图谱；推荐
Abstract
Comparing with the traditional neural network, the neural network based on knowledge graph (KG) takes the graph as the input in the recommender system, which can combine the node information and topology for prediction. However, the existing methods rarely consider the symmetry relationship in KG and the disappearance of the gradient. A Neural Network Recommender Algorithm Based on Bidirectional Graph Attention (BGANR) is proposed. Firstly, the graph neural network with the symmetrical attention mechanism and the bidirectional translation model are combined to embed the KG information. Then, the dynamic activation function is used to avoid increasing of calculation and overfitting. Empirical results on two benchmark datasets demonstrate our model outperforms state-of-the-art methods.
Key words： Bidirectional embedding; Attention mechanism; Knowledge graph; Recommender
参考文献 References
[1] HE X, LI L, et.al. Neural Collaborative Filtering[C]. //In WWW, 2017. 173–182.
[2] CHEN J, et.al. Attentive Collaborative Filtering: Multimedia Recommender with Item- and Component-Level Attention[C]. // In SIGIR, 2017. 335–344.
[3] WANG, H., et al . Knowledge Graph Convolutional Networks for Recommender Systems. arXiv (2019).
[4] ZHOG G, SONG C, et al. Deep Interest Network for Click-Through Rate Prediction[J]. arXiv preprint arXiv:1706.06978,2017.
[5] SHI C, HU B, et al. Heterogeneous Information Net-work Embedding for Recommendation[J].IEEE Transactions on Knowledge & Data Engineering,2017:125-136.
[6] CHEN J, et.al. Attentive Collaborative Filtering: Mul-timedia Recommender with Item- and Component-Le-vel Attention[C]. // In SIGIR, 2017. 335–344.
[7] WANG X, WANG D, et al . Explainable Reasoning over Knowledge Graphs for Recommendation[C]. //In AAAI 2008:565-668.
[8] RENDLE S, FREUDENTHALER C et al. BPR: Bayesian personalized ranking from implicit feedback[C]//UAI 2009 Proceedings of the Twenty-Fifth Conference on Uncertainty in Artificial Intelligence,2009. Montreal, QC, Canada: AUAI Press, 2009,452-461.
[9] ZHANG F, YUAN N, et al. Collaborative Knowledge Base Embedding for Recommender Systems[C]// the 22nd ACM SIGKDD International Conference. ACM, 2016, 236-245.
[10] ZHANG Y,CHEN X, WANG P. Learning over know-ledge-base embeddings for recommendation, arXiv preprint arXiv:1803.06540, 2018.