计算机工程 ›› 2019, Vol. 45 ›› Issue (9): 176-182.doi: 10.19678/j.issn.1000-3428.0051339

• 人工智能及识别技术 • 上一篇    下一篇

基于注意力机制与评论文本深度模型的推荐方法

黄文明a,b, 卫万成a, 张健a, 邓珍荣a,b   

  1. 桂林电子科技大学 a. 计算机与信息安全学院;b. 广西可信软件重点实验室, 广西 桂林 541004
  • 收稿日期:2018-04-24 修回日期:2018-08-22 出版日期:2019-09-15 发布日期:2019-09-03
  • 作者简介:黄文明(1963-),男,教授,主研方向为人工智能、大数据处理、图形图像处理;卫万成、张健,硕士研究生;邓珍荣,研究员。
  • 基金项目:
    广西自然科学基金(2018GXNSFAA138132);广西高校云计算与复杂系统重点实验室项目(yf17106);桂林电子科技大学研究生教育创新计划(2018YJCX55)。

Recommendation Method Based on Attention Mechanism and Review Text Deep Model

HUANG Wenminga,b, WEI Wanchenga, ZHANG Jiana, DENG Zhenronga,b   

  1. a. School of Computer Science and Information Security;b. Guangxi Key Laboratory of Trusted Software, Guilin University of Electronic Technology, Guilin, Guangxi 541004, China
  • Received:2018-04-24 Revised:2018-08-22 Online:2019-09-15 Published:2019-09-03

摘要: 传统推荐系统依赖人工进行规则设计和特征提取,对评论文本内容的特征和隐信息的提取能力有限。针对该问题,融合注意力机制并基于深度学习对推荐系统进行改进,提出一种对评论文本深度建模的推荐方法。使用词嵌入模型表达数据集评论中的语义,引入注意力机制对输入内容进行重新赋权,通过并行的卷积神经网络挖掘用户和项目评论数据中的隐含特征,将两组特征耦合输入并采用因子分解机进行评分预测,得到推荐结果。实验结果表明,该方法可有效提高推荐准确率,均方误差较DeepCoNN方法提升2%以上。

关键词: 推荐系统, 特征提取, 注意力机制, 卷积神经网络, 因子分解机

Abstract: The traditional recommendation system relies on manual rule design and feature extraction,resulting in insufficient extraction of the features and implicit information of the review text content.Aiming at this problem,a recommendation method for deep modeling of review text is proposed combining the attention mechanism and the improved recommendation system based on deep learning.The word embedding model is used to express the semantics in dataset comments,the attention mechanism is introduced to re-weight the input content,and the hidden features in the user and project comment data are mined through the parallel convolutional neural network.The two sets of features are coupled and input and scored by a factorization machine to obtain the recommendation results.Experimental results show that the proposed algorithm can efficiently improve the recommendation precision,and the mean square error is improved by more than 2% than that of the DeepCoNN algorithm.

Key words: recommendation system, feature extraction, attention mechanism, Convolution Neural Network(CNN), factorization machine

中图分类号: