基于评论文本的自适应特征提取推荐研究  

Research on Review Text Based Self-adaptive Feature Extraction Recommendation

作　　者：胡海星 王宜贵 袁卫华[1] 张志军[1] 秦倩倩 HU Hai-xing;WANG Yi-gui;YUAN Wei-hua;ZHANG Zhi-jun;QIN Qian-qian(School of Computer Science and Technology,Shandong Jianzhu University,Ji’nan 250101,China)

机构地区：[1]山东建筑大学计算机科学与技术学院,山东济南250101

出　　处：《软件导刊》2022年第2期1-7,共7页Software Guide

基　　金：国家自然科学基金项目(62177031);山东省自然科学基金项目(ZR2021MF099);山东省教学改革研究项目(2021)。

摘　　要：现有评论文本推荐方法多使用静态词向量技术获取评论嵌入,但单词多义性会对语义理解产生偏差,且特征拼接策略无法平衡用户和商品特征对推荐结果的影响。为此,提出了基于评论文本的自适应特征提取推荐模型。该模型使用动态词嵌入预训练模型BERT解决多义性问题,结合Bi-GRU与注意力机制的双向特征提取增强特征表达能力,并以自适应特征拼接机制平衡用户和商品特征在交互时的贡献程度。实验结果表明,该模型在6个亚马逊数据集上均方误差值最低为0.678,相比最优基准模型性能平均提高了2.42%,有效改善评论文本中单词多义性问题对推荐结果的影响,自适应特征拼接机制有效平衡了用户和商品特征各自的重要程度,提高了预测评分精度。Existing recommendation methods based on review text obtain review embeddings through static word vector technologies,and the ambiguity of words leads to the problem of deviations in semantic understanding. Besides,feature splicing strategies cannot balance the impact of user feature and item feature on recommendations. To solve the above problems,propose a review text based adaptive feature extraction recommendation model. First,the model uses the dynamic word embedding technology-BERT as its pretraining model. Next,the model combines Bi-GRU with the attention mechanism for bidirectional feature extraction to fully extract reviews features. Finally,an adaptive feature splicing mechanism is proposed to balance the contribution of user features and item features during interactions. The experimental results show that the minimum mean square error value of the model on the 6 Amazon data sets is0.678,which is an average increase of 2.42% compared to the performance of the best baseline. The model can not only reduce the impact of word ambiguity in the review text on the recommendation,but also the adaptive feature splicing mechanism can balance the importance of users and products,and improve the accuracy of prediction rating.

关 键 词：推荐系统 深度学习 自适应特征提取 评论文本 注意力机制 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]