基于深度学习和词汇相似度的个性化新闻推荐系统设计  

Design of personalized news recommendation system based on lexical similarity and deep learning

作　　者：江涛 Jiang Tao(Information Technology School,Guangdong Polytechnic College,Zhaoqing 526000,China)

机构地区：[1]广东理工学院信息技术学院,广东肇庆526000

出　　处：《无线互联科技》2021年第7期47-49,共3页Wireless Internet Technology

基　　金：广东理工学院质量工程项目,项目名称:基于深度学习的以词汇相似度为基础的个性化新闻推荐系统分析,项目编号:JXGG202053。

摘　　要：对网络上庞大的新闻资讯,如何发展一个个性化的新闻推荐系统,自动地推荐使用者感兴趣的新闻,是一个备受重视的课题。文章提出一个个性化新闻推荐系统,此系统将建立一个新闻本体,并通过深度学习计算使用者偏好,以此达到推荐个性化新闻的目的。此新闻本体以分析新闻的词汇为基础,并参考专家的分类。其中,每个类别包含特定数量的代表性词汇,而这些词汇以时事新闻进行TF-IDF统计而得。对每一则新闻,系统将计算该则新闻所包含的词汇与新闻本体中代表性词汇的相似度,定义为新闻的特征向量,并将此特征向量输入多层次类神经网络进行深度学习计算得出新闻推荐值。实验结果显示,相较于随机推荐,文章所提出的方法可以较大地提升推荐成功的比率,神经网络将由推荐值来判断是否推荐给使用者,若是使用者未点击阅读此新闻,判断为使用者不喜欢此篇新闻,神经网络将会进行修正,使之越来越接近真实的使用者偏好。For the huge news information on the Internet,how to develop a personalized news recommendation system to automatically recommend the news that users are interested in is a highly valued topic.This paper proposes a personalized news recommendation system,which will build a news ontology and calculate user preferences through deep learning,so as to achieve the purpose of learning to recommend personalized news.This news ontology is based on analyzing news vocabulary and referring to the classification of experts,in which each category contains a specific number of representative vocabulary,which is obtained by TF-IDF statistics of current news.For each news,the system will calculate the similarity between the words contained in the news and the representative words in the news ontology,and define it as the feature vector of the news,and then input this feature vector into the multi-level neural network for deep learning to calculate the news recommendation value.Experimental results show that compared with random recommendation,the method proposed in this paper can greatly improve the success rate of recommendation.The neural network will judge whether to recommend it to the user by the recommended value.If the user doesn’t click to read the news and judges that the user doesn’t like the news,the neural network will revise it to make it more and more close to the real user preference.

关 键 词：使用者偏好 新闻推荐 深度学习 TF-IDF 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术] TP18[自动化与计算机技术—计算机科学与技术]