赞助商式广告推荐系统中数据稀疏问题的研究  

Research About Data Sparseness in Sponsored Search Advertising

作　　者：杨燕[1] 霍晓骏 贺樑[1] 

机构地区：[1]华东师范大学信息科学技术学院,上海200241

出　　处：《小型微型计算机系统》2015年第9期1993-1997,共5页Journal of Chinese Computer Systems

基　　金：国家科技支撑项目(2012BAH93F03)资助;上海市科委项目(13511506201)资助

摘　　要：广告推荐系统是当前针对网络广告的研究热点.在广告推荐系统中,一般利用使用基于邻域的协同过滤产生推荐,但是这种算法对数据的稀疏性有着一定要求,不适合对稀疏性太高的数据使用协同过滤.本文利用矩阵分解算法解决数据稀疏的问题,并在腾讯搜搜广告日志数据上进行实验.实验结果表明,本文提出的矩阵分解广告推荐算法弥补了数据稀疏带来的弊端,能够取得更高的准确率,召回率以及F指标,对稀疏的数据有着更高的抵抗性.Advertising recommendation system is the current hotspot for online advertising. For an ad recommendation system,neighborhood-based collaborative filtering can be a good choice. How ever,this kind of algorithm has a certain requirements of data sparseness,if the data is too sparse,it is not suitable for neighborhood-based collaborative filtering. In this article w e use matrix factorization to fix the problem of data sparseness. An experiment on tencent soso data about ad logs show s that our advertising matrix factorization recommendation algorithm has a better recommendation result of higher recall,precision and F-measure than baseline on sparse data.

关 键 词：广告推荐 矩阵分解 协同过滤 赞助商式广告 稀疏 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]