基于协同过滤技术推荐系统的探究  

Research on Recommender Systems Based on Collaborative Filtering

作　　者：江水[1] JIANG Shui(The 32nd Research Institute of China Electronics Technology Group Corporation,Shanghai 201808,China)

机构地区：[1]中国电子科技集团公司第三十二研究所,上海201808

出　　处：《计算机技术与发展》2021年第11期1-7,共7页Computer Technology and Development

基　　金：国家自然科学基金资助项目(61402497)。

摘　　要：基于数据挖掘技术的推荐系统在互联网和大数据时代无处不在,它能够根据用户的兴趣和行为特征,为其推荐感兴趣的项目,具有深远的商业价值。协同过滤技术综合考虑用户的历史特征和用户之间的相似关系,为用户进行预测推荐,已经广泛应用于现有的推荐系统中。文中阐述了协同过滤技术的概念内容、前提假设、数据采集与过滤模式、数据结构、评价标准和算法分类,并着重分析对比了基于存储、基于模型和关联规则等三种推荐系统算法的不同实现方式及其优缺点。在此基础上,编程实现了向量相似性算法、个性诊断算法、奇异值分解算法和关联算法等典型常用算法,并在Movielens数据集和稀疏成绩单数据库上进行了验证与评估,实验结果可以为设计人员提供指导依据。Recommender systems based on data mining technology have been widely used in the era of Internet and big data.They can recommend relevant items to users according to their interests and behaviors,which contributes a deep commercial value.Collaborative filtering belongs to the most widely used recommender systems,which can implement prediction and recommendation using the history pattern of users and their similarity.We systematically introduce the concept,assumption,data collection and filtering mode,data structure,evaluation criteria and classification of collaborative filtering,analyze and compare different implementation and advantage and disadvantage of memory-based,model-based and association rules.On the basis,we have programmed several typical algorithms,such as vector similarity algorithm,personality diagnosis algorithm,singular value decomposition algorithm and correlation analysis algorithm,which are verified and evaluated on Movielens dataset and sparse report card database.The experimental results can provide instructions for system designers.

关 键 词：推荐系统 数据挖掘 协同过滤 大数据 个性诊断 

分 类 号：TP302[自动化与计算机技术—计算机系统结构]