计算机工程 ›› 2017, Vol. 43 ›› Issue (12): 173-178.doi: 10.3969/j.issn.1000-3428.2017.12.032

• 人工智能及识别技术 • 上一篇    下一篇

基于降噪自编码器网络与词向量的信息推荐方法

郭喻栋 1,郭志刚 1,席耀一 2   

  1. (1.解放军信息工程大学 信息系统工程学院,郑州 450002; 2.解放军外国语学院,河南 洛阳 471003)
  • 收稿日期:2016-11-22 出版日期:2017-12-15 发布日期:2017-12-15
  • 作者简介:郭喻栋(1991—),男,硕士研究生,主研方向为自然语言处理;郭志刚,副教授;席耀一,讲师。
  • 基金项目:
    国家社会科学基金“网上舆论斗争系统建模与应对策略研究”(14BXW028)。

Information Recommendation Method Based on Denoising Auto-encoder Network and Word Vector

GUO Yudong  1,GUO Zhigang  1,XI Yaoyi  2   

  1. (1.College of Information System Engineering,PLA Information Engineering University,Zhengzhou 450002,China; 2.PLA Foreign Language College,Luoyang,Henan 471003,China)
  • Received:2016-11-22 Online:2017-12-15 Published:2017-12-15

摘要: 基于降噪自编码器的推荐方法缺乏对项目共现关系的分析能力,且模型存在参数冷启动问题。为此,基于降噪自编码器网络与词向量,提出一种信息推荐方法。将用户看作文档,将用户评价过的项目看作词语,以此构建训练语料。利用词向量模型对语料进行训练,得到隐含上下文信息的项目向量。将所有项目向量作为初始权重构建降噪自编码器神经网络,训练得到模型参数。通过模型预测用户评分完成top-N推荐。在标准数据集上的实验结果表明,该方法能提高推荐准确率,训练速度优于降噪自编码、奇异值矩阵分解和协同过滤推荐方法。

关键词: 信息推荐, 神经网络, 降噪自编码器, 词向量, 参数冷启动

Abstract: The recommendation method based on Denoising Auto-encoder(DAE)lacks of items co-occurrence analysis ability,and the model exists the problem of cold start of parameters.In order to solve these problems,this paper proposes an information recommendation method based on DAE network and word vector.A training corpus is built by mapping users into documents and mapping items into words.The word vector model is used to train the corpus to generate item vectors which contain implicit context information.All item vectors are used as the initial weights to reconstruct the DAE neural network,and the model parameters are obtained through training.The model is used to predict ratings to complete top-N recommendation.Experimental results on standard datasets show that the proposed method improves the accuracy of recommendation,and the training speed is better than that of DAE,Singular Value Decomposition(SVD) and Collaborative Filtering(CF) methods.

Key words: information recommendation, neural network, Denoising Auto-encoder(DAE), word vector, cold start of parameter

中图分类号: