推荐系统冷启动问题解决策略研究  

Research on Solution of Solving Cold Start Problem in Recommender Systems

作　　者：乔雨[1] 李玲娟[1] 

机构地区：[1]南京邮电大学计算机学院,江苏南京210003

出　　处：《计算机技术与发展》2018年第2期83-87,共5页Computer Technology and Development

基　　金：国家自然科学基金(61302158;61571238)

摘　　要：推荐系统利用机器学习技术进行信息过滤,快速准确地定位用户需要的信息,并且能够预测用户对目标项目的喜好程度。由于新用户与新项目的存在,传统的推荐系统在缺少数据信息的情况下面临着冷启动问题的挑战,导致系统无法为用户产生准确的推荐。分析冷启动产生的原因,阐述解决冷启动问题的意义,从是否考虑冷启动类型等方面对目前推荐系统冷启动问题的研究成果进行分类总结,并尝试给出冷启动问题未来的研究重点与难点。目前较为普遍的处理方式是将多种数据源与多种推荐方法进行混合使用,从而提高系统推荐的准确度与效率,但是仍然存在着如在收集用户各类信息的同时如何保护个人隐私、如何建立推荐系统的效用评价等难点问题。Recommendation systems apply machine learning techniques to filter and locate information accurately,and can predict whether auser would like a given resource. Traditional collaborative filtering systems have to deal with the cold-start problems as new users and itemsare always present,which fail to produce the accurate recommendation for users. In this paper we first illustrate the causes and the significances of solving the cold-start problems according to the current achievements in research,and then summarize the existing algorithms andcompare the performance of them. Finally,we try to give the difficulties and future directions of recommender system. It was found that themost popular way is to have mixed data sources and algorithms to improve the accuracy and efficiency of recommender system at present,butthere still have been some difficulty like how to protect personal privacy during getting users information or how to establish the performance evaluation of recommendation systems.

关 键 词：推荐系统 协同过滤 用户冷启动 项目冷启动 解决策略 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]