协同过滤推荐系统综述

赵俊逸; 庄福振; 敖翔; 何清; 蒋慧琴; 马岭

引用本文：

赵俊逸,庄福振,敖翔,何清,蒋慧琴,马岭.协同过滤推荐系统综述[J].信息安全学报,2021,6(5):17-34 [点击复制]
ZHAO Junyi,ZHUANG Fuzhen,AO Xiang,HE Qing,JIANG Huiqin,MA Ling.Survey of Collaborative Filtering Recommender Systems[J].Journal of Cyber Security,2021,6(5):17-34 [点击复制]

本文已被：浏览 15288次下载 10909次	码上扫一扫！
协同过滤推荐系统综述
赵俊逸^1,2, 庄福振³, 敖翔^1,2, 何清^1,2, 蒋慧琴⁴, 马岭⁴
0 字体:加大+\|默认\|缩小-
(1.郑州大学河南先进技术研究院, 郑州中国 450003;2.中国科学院计算技术研究所智能信息处理重点实验室北京中国 100190;3.北京航空航天大学人工智能研究院北京中国 100191;4.郑州大学信息工程学院郑州中国 450003)

摘要:

随着互联网和信息计算的飞速发展，衍生了海量数据，我们已经进入信息爆炸的时代。网络中各种信息量的指数型增长导致用户想要从大量信息中找到自己需要的信息变得越来越困难，信息过载问题日益突出。推荐系统在缓解信息过载问题中起着非常重要的作用，该方法通过研究用户的兴趣偏好进行个性化计算，由系统发现用户兴趣进而引导用户发现自己的信息需求。目前，推荐系统已经成为产业界和学术界关注、研究的热点问题，应用领域十分广泛。在电子商务、会话推荐、文章推荐、智慧医疗等多个领域都有所应用。传统的推荐算法主要包括基于内容的推荐、协同过滤推荐以及混合推荐。其中，协同过滤推荐是推荐系统中应用最广泛最成功的技术之一。该方法利用用户或物品间的相似度以及历史行为数据对目标用户进行推荐，因此存在用户冷启动和项目冷启动问题。此外，随着信息量的急剧增长，传统协同过滤推荐系统面对数据的快速增长会遇到严重的数据稀疏性问题以及可扩展性问题。为了缓解甚至解决这些问题，推荐系统研究人员进行了大量的工作。近年来，为了提高推荐效果、提升用户满意度，学者们开始关注推荐系统的多样性问题以及可解释性等问题。由于深度学习方法可以通过发现数据中用户和项目之间的非线性关系从而学习一个有效的特征表示，因此越来越受到推荐系统研究人员的关注。目前的工作主要是利用评分数据、社交网络信息以及其他领域信息等辅助信息，结合深度学习、数据挖掘等技术提高推荐效果、提升用户满意度。对此，本文首先对推荐系统以及传统推荐算法进行概述，然后重点介绍协同过滤推荐算法的相关工作。包括协同过滤推荐算法的任务、评价指标、常用数据集以及学者们在解决协同过滤算法存在的问题时所做的工作以及努力。最后提出未来的几个可研究方向。

关键词: 推荐系统协同过滤稀疏性深度学习

DOI：10.19363/J.cnki.cn10-1380/tn.2021.09.02

投稿时间：2021-04-30修订日期：2021-08-08

基金项目:本文得到国家重点研发计划课题（No.2017YFB1002104），国家自然科学基金（No.61976204，No.U1811461），中科院青年创新促进会支持

Survey of Collaborative Filtering Recommender Systems

ZHAO Junyi^1,2, ZHUANG Fuzhen³, AO Xiang^1,2, HE Qing^1,2, JIANG Huiqin⁴, MA Ling⁴

(1.Henan Institute of Advanced Technology, Zhengzhou University, Zhengzhou 450003, China;2.Key Lab of Intelligent Information Processing of Chinese Academy of Sciences (CAS), Institute of Computing Technology, CAS, Beijing 100190, China;3.Institute of Artificial Intelligence, Beihang University, Beijing 100191, China;4.School of Information Engineering, Zhengzhou University, Zhengzhou 450003, China)

Abstract:

With the rapid development of Internet and information computing, we have entered the era of information explosion. The exponential growth of all kinds of information in the network makes it more and more difficult for users to find the information they need from a large amount of information, and the problem of information overload is becoming increasingly prominent. Recommender system plays a very important role in alleviating the problem of information overload. This method studies the user's interest preference for personalized calculation, and the system finds the user's interest, and then guides the user to find their own information needs. At present, recommender system has become a hot issue in industry and academia. It has a wide range of applications, such as e-commerce, conversation recommendation, article recommendation, intelligent medical and so on. Traditional recommendation algorithms mainly include content-based recommendation, collaborative filtering recommendation and hybrid recommendation. Among them, collaborative filtering recommendation is one of the most widely used and successful technologies in recommender system. This method uses the similarity between users or items and historical behavior data to recommend target users, so there are problems of user cold start and project cold start. In addition, with the rapid growth of information, traditional collaborative filtering recommender system will encounter serious data sparsity and scalability problems in the face of the rapid growth of data. In order to alleviate or even solve these problems, recommender system researchers have done a lot of work. In recent years, in order to improve the recommendation effect and user satisfaction, scholars began to pay attention to the diversity and interpretability of recommender system. Because deep learning method can learn an effective feature representation by discovering the nonlinear relationship between users and items in the data, it has attracted more and more attention of recommender system researchers. The current work is mainly to use scoring data, social network information and other areas of information and other auxiliary information, combined with deep learning, data mining and other technologies to improve the recommendation effect and improve user satisfaction. In this regard, this paper first summarizes the recommender system and traditional recommendation algorithm, and then focuses on the related work of collaborative filtering recommendation algorithm. It includes the task of collaborative filtering recommendation algorithm, evaluation index, common data sets and the work and efforts of scholars in solving the problems of collaborative filtering algorithm. Finally, several possible research directions in the future are proposed.

Key words: recommender system collaborative filtering sparsity deep learning