基于注意力机制的神经网络贝叶斯群组推荐算法  

A NEURAL NETWORK BAYESIAN GROUP RECOMMENDATION ALGORITHM BASED ON ATTENTION MECHANISM

作　　者：李诗文 潘善亮[1] Li Shiwen;Pan Shanliang(Faculty of Electrical Engineering and Computer Science,Ningbo University,Ningbo 315211,Zhejiang,China)

机构地区：[1]宁波大学信息科学与工程学院,浙江宁波315211

出　　处：《计算机应用与软件》2020年第5期287-292,共6页Computer Applications and Software

基　　金：浙江省公益性技术应用研究计划项目(2017C33001)。

摘　　要：针对群组推荐算法中矩阵稀疏和群组偏好融合问题,提出一种基于注意力机制的神经网络贝叶斯群组推荐算法(ANBGR)。利用群组和项目的交互记录,生成群组对于正项目和负项目的偏好三元组,然后将群组中的用户和项目映射到高维空间,生成用户和项目的潜在矩阵信息。注意力模型使用不同权重下用户偏好聚合和用户偏好交互的策略,隐藏层使用多层感知机网络结构生成群组非线性高维特征。最终基于贝叶斯理论优化算法实现群组对于项目的偏好预测,实验结果表明,该方法相对于对比实验算法有更好的推荐结果。Aiming at the problem of matrix sparseness and group preference fusion in group recommendation algorithm,we propose a neural network Bayesian group recommendation algorithm(ANBGR)based on attention mechanism.It used the interaction records of groups and projects to generate the group preference triples for positive and negative projects,and then mapped users and projects in the group to high-dimensional space,so as to generate potential matrix information of users and projects.The attention model used the strategy of user preference aggregation and user preference interaction under different weights.The hidden layer used the multi-layer perceptron network structure to generate group nonlinear high-dimensional features.Finally,based on Bayesian theory optimization algorithm,the group s preference prediction for the project was realized.The experimental results show that our method has better recommendation results than the comparison algorithm.

关 键 词：群组推荐算法 注意力模型 贝叶斯排序算法 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]