基于多准则决策和深度神经网络的电子商务推荐系统  

E-commerce Recommendation System Based on Multi-Criteria Decision Making and Deep Neural Networks

作　　者：韩晓路[1] 周湘贞[2] HAN Xiao-Lu;ZHOU Xiang-zhen(Modern Business School,Anhui Vocational and Technical College,Anhui,Hefei 230001;Department of Information Engineering,Shengda University,Henan,Zhengzhou 451191)

机构地区：[1]安徽职业技术学院现代商务学院,安徽合肥451191 [2]郑州升达经贸管理学院信息工程学院,河南郑州230001

出　　处：《贵阳学院学报（自然科学版）》2024年第3期69-73,91,共6页Journal of Guiyang University：Natural Sciences

基　　金：安徽省高校科学研究重大项目(哲学社科)(项目编号:2023AH040186);河南省2022年度科技厅科技攻关项目(项目编号:222102210290)。

摘　　要：推荐系统在电子商务门户中发挥着重要作用。现有推荐算法通常基于单个评分标准进行推荐商品排序,未考虑到从不同标准的用户偏好上对用户和商品特征进行综合建模。为此,提出了基于多准则决策(MCD)和深度神经网络(DNN)的电商产品推荐系统。首先,设计上下文感知的DNN模型,从不同标准出发进行评分预测,并通过聚合函数得到综合评分预测值。其后,通过结合残差卷积神经网络(CNN)和双向长短时记忆(Bi-LSTM)的混合模型,基于用户评论预测用户对商品的情感倾向。最后,将评分预测与情感倾向结合,实现准确的商品推荐。实验结果表明,所提方法在亚马逊电商产品数据集上进行商品推荐的平均绝对误差(MAE)和均方根误差(RMSE)分别为0.953和1.129,优于其他比较方法,证明在推荐系统中结合MCD和情感分析,能够有效提高推荐系统性能。Recommendation systems play a crucial role in e-commerce portals.Existing recommendation algorithms typ-ically rank products based on a single rating criterion,neglecting the comprehensive modeling of user and product char-acteristics from dfferent criteria of user preferences.Therefore,an e-commerce product recommendation system based on multi-criteria decision(MCD)and deep neural networks(DNN)is proposed.First,a context-aware DNN model is designed to predict ratings from different criteria and obtain an aggregated rating prediction through an aggregation function.Subsequently,a hybrid model combining residual convolutional neural networks and bidirectional long short-term memory(Bi-LSTM)is employed to predict user sentiment towards products based on user reviews.Finally,the rating predictions and sentiment tendencies are integrated to achieve accurate product recommendations.Experimental results indicate that the proposed method achieves a mean absolute error(MAE)of 0.953 and a root mean square error(RMSE)of 1.129 on the Amazon e-commerce product dataset,outperforming other comparative methods.This dem-onstrates that integrating MCD and sentiment analysis in recommendation systems can effectively enhance their perform-ance.

关 键 词：电子商务 推荐系统 多准则决策 深度神经网络 情感分析 双向长短时记忆 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]