基于知识图谱的长短期序列推荐算法

A long and short-term sequence recommendation algorithm based on knowledge graph

现有的部分序列推荐算法较少关注用户短期兴趣随时间变化的问题,从而导致推荐的精度不够理想,且在用户兴趣转变的可解释性上有待提高。据此,提出了一种基于知识图谱的长短期序列推荐算法(KGLSR)。将交互历史划分为长期和短期行为序列后,结合卷积神经网络与注意力机制进行长期兴趣的特征重构,并引入知识图谱与图注意力更新用户的短期偏好,最后实现自适应聚合。经验证,该模型在3类真实场景下的数据集中以HR、MRR和NDCG为评价指标的表现均优于对比实验中的主流基线模型。

Existing partial sequence recommendation algorithms pay less attention to the problem of users short-term interest over time,which leads to less-than-ideal recommendation accuracy and to-be-improved interpretability of users interest shift.Therefore,a long-short-term sequence recommendation algorithm based on knowledge graph,namely KGLSR,is proposed.After dividing the interaction history into long and short-term behavioral sequences,it combines the convolutional neural network(CNN)with the attention mechanism for feature reconstruction of long-term interests,and introduces the knowledge graph with GAT to update users short-term preferences.Thus,the adaptive aggregation is realized.The comparison experiments demonstrate that the proposed model outperforms the mainstream baseline model in terms of evaluation metrics of HR,MRR and NDCG,on datasets of 3 types of real scenarios.