融合偏好传播的多任务推荐模型研究

Research on Multi-task Recommendation Model Fused with Preference Propagation

针对知识图谱可以有效地从多源异构数据中还原出实体的三元组关系,却不利于推荐任务且采用单任务学习又很难挖掘数据潜在关联的问题,提出一种融合偏好传播的多任务推荐模型(MAPKR)。首先,利用涟漪传播从知识图谱中提取用户的偏好特征集;其次,依据相似近邻结构实现潜在特征共享,经交叉压缩单元提取项目和实体的高阶特征表示;最后,以多任务学习交替训练推荐模块和知识图谱嵌入模块,将提取的特征向量经归一化内积操作后进行预测、推荐。在3个公开数据集上进行实验并与5个基线模型进行比较。与MKR和RippleNet相比,在MovieLens-1M数据集上,AUC和ACC分别提高了0.68%、0.31%和0.77%、0.54%;在Book-Crossing上,AUC和ACC分别提高了3.48%、2.66%和4.51%、7.21%;在Last.FM上,AUC和ACC分别提高了3.44%、6.25%和2.70%、2.62%。实验结果表明,提出的模型与MKR、RippleNet等其他基线模型相比推荐性能良好。

To address the problem that the knowledge graph can effectively reduce the triadic relationships of entities from multi-source het-erogeneous data,but is not conducive to recommendation tasks and it is difficult to explore the potential association relationships of data using single-task learning,a multi-task recommendation model with fused preference propagation(MAPKR)is proposed.Firstly,the user's prefer-ence feature set is extracted from the knowledge graph using ripple propagation;secondly,the potential features are shared based on the simi-lar nearest neighbor structure,and the higher-order feature representations of items and entities are extracted by cross-compression units;fi-nally,the recommendation module and the knowledge graph embedding module are trained alternately with multi-task learning,and the ex-tracted feature vectors are predicted and recommended after normalized inner product operation.Experiments are conducted on three publicly available datasets and compared with five baseline models.Compared with MKR and Ripple Net,the AUC and ACC are improved by 0.68%,0.31%and 0.77%,0.54%on MovieLens-1M dataset;3.48%,2.66%and 4.51%,7.21%on Book-Crossing,respectively;on Last.FM,AUC and ACC improved by 3.44%,6.25%and 2.70%,2.62%,respectively.The experimental results show that the proposed model has good recommendtion performance compared with other baseline models such as MKR and RippleNet.