基于双重注意力机制和时间因子的深度推荐模型  

A Deep Recommendation Model Based on Dual AttentionMechanism and Time Factor

作　　者：吴定谕 周从华[1] 单田华[1] 刘志锋[1] WU Dingyu;ZHOU Conghua;SHAN Tianhua;LIU Zhifeng(School of Computer Science and Telecommunication Engineering,Jiangsu University,Zhenjiang 212013)

机构地区：[1]江苏大学计算机科学与通信工程学院,镇江212013

出　　处：《计算机与数字工程》2023年第3期679-685,共7页Computer & Digital Engineering

基　　金：江苏省重点研发计划(社会发展)项目(编号:BE2016630,BE2017628)资助。

摘　　要：现阶段大多数基于评论文本的推荐模型没有从多个视角充分挖掘用户评论的价值,忽略了评论文本在不同层面的重要度信息。基于此,论文提出一种基于双重注意力机制和时间因子的深度推荐模型DATCoNN。该模型使用并行的卷积神经网络结合两层注意力层分别挖掘单词层面和评论层面的重要度信息,然后使用时间因子进一步拟合用户对项目兴趣度的变化情况,最终采用因子分解机实现评分预测。模型在Amazon的三组不同领域数据集上进行对比实验评估,发现论文提出的推荐模型性能最优,同时该模型具有较好的可解释性。At present,most of the recommendation models based on review text has not fully mined the value of reviews from multiple perspectives.They ignore that the review contains a lot of information at different levels.Based on these reasons,this paper proposes a deep recommendation model DATCoNN based on dual attention mechanism and time factor.The model uses parallel con⁃volutional neural network combined with two attention layers to mine the importance of review at word level and review level respec⁃tively,the time factor is used to fit the change of user's interest in items.Finally,the model uses factorization machine to obtain the recommendation results.In this paper,the experimental comparison and analysis is carried out on three groups of datasets in differ⁃ent fields provided by Amazon.The experimental results show that the proposed model obtain the best performance.Moreover,the proposed recommendation model has better interpretability.

关 键 词：推荐系统 注意力机制 卷积神经网络 时间因子 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]