融合注意力LSTM的协同过滤推荐算法  

Collaborative Filtering Based on Attention LSTM

作　　者：罗洋 夏鸿斌[1,2] 刘渊 LUO Yang;XIA Hongbin;LIU Yuan(School of Digital Media,Jiangnan University,Wuxi,Jiangsu 214122,China;Jiangsu Key Laboratory of Media Design and Software Technology,Wuxi,Jiangsu 214122,China)

机构地区：[1]江南大学数字媒体学院,江苏无锡214122 [2]江苏省媒体设计与软件技术重点实验室,江苏无锡214122

出　　处：《中文信息学报》2019年第12期110-118,共9页Journal of Chinese Information Processing

基　　金：国家科学支撑计划课题(2015BAH54F01);国家自然科学基金(61672264)

摘　　要：针对传统协同过滤算法难以学习深层次用户和项目的隐表示,以及对文本信息不能充分提取单词之间的前后语义关系的问题,该文提出一种融合辅助信息与注意力长短期记忆网络的协同过滤推荐模型。首先,附加堆叠降噪自编码器利用评分信息和用户辅助信息提取用户潜在向量;其次,基于注意力机制的长短期记忆网络利用项目辅助信息来提取项目的潜在向量;最后,将用户与项目的潜在向量用于概率矩阵分解中,从而预测用户偏好。在两个真实数据集MovieLens-100k和MovieLens-1M上进行实验,采用RMSE和Recall指标进行评估。实验结果表明,该模型与其他相关推荐算法相比在推荐性能上有所提升。To better capture the implicit representation for the user and the item,and the semantic relationship between words,a collaborative filtering recommendation model combining auxiliary information and attention LSTM is proposed.Firstly,the additional stacked denoising autoencoder is applied to extract the user potential vector from the scoring information and the user auxiliary information.Secondly,the LSTM with attention mechanism is utilized to extract the potential vector of the item from the item auxiliary information.Finally,the user and the item potential vectors are used in the probability matrix factorization to predict user preferences.Experiments on two real data sets,movielens-100 k and movielens-1 m,show that the proposed model has improved performance compared with other recommendation algorithms.

关 键 词：注意力机制 长短期记忆网络 推荐系统 附加堆叠降噪自编码器 协同过滤 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]