基于概率图模型的互联网广告点击率预测

发布时间：2018-06-08 11:59

本文选题：计算广告 + 点击率　；参考：《华东师范大学学报(自然科学版)》2013年03期

【摘要】：点击率预测可以提高用户对所展示互联网广告的满意度,支持广告的有效投放,是针对用户进行广告的个性化推荐的重要依据.对于没有历史点击记录的用户,仍需对其推荐广告,预测所推荐广告的点击率.针对这类用户,以贝叶斯网这一重要的概率图模型,作为不同用户之间广告搜索行为的相似性及其不确定性的表示和推理框架,通过对用户搜索广告的历史记录进行统计计算,构建反映用户间相似关系的贝叶斯网,进而基于概率推理机制,定量度量没有历史点击记录的用户与存在历史点击记录的用户之间的相似性,从而预测没有历史点击记录的用户对广告的点击率,为广告推荐提供依据.通过建立在KDD Cup 2012-Track 2的Tencent CA训练数据集上的实验,测试了方法的有效性.
[Abstract]:The prediction of click rate can improve the users' satisfaction with the displayed Internet ads and support the effective placement of advertisements. It is an important basis for personalized recommendation of advertisements for users. For users who have no history click record, they still need to recommend ads and predict the click rate of the recommended ads. For this kind of users, Bayesian network, an important probability graph model, is used as the representation and reasoning framework of the similarity and uncertainty of advertisement search behavior among different users. A Bayesian network is constructed to reflect the similarity between users, and then based on probabilistic reasoning mechanism, the similarity between users without historical click records and those with historical click records is quantitatively measured. Thus, it predicts the click rate of the users who have no history click record, and provides the basis for advertising recommendation. The validity of the method is tested by the experiment on the Tencent CA training data set of KDD Cup 2012-Track2.
【作者单位】：云南大学信息学院计算机科学与工程系;
【基金】：国家自然科学基金(61063009,61163003,61232002) 国家教育部博士点基金新教师类课题(20105301120001) 云南省中青年学术与技术带头人后备人才培养计划(2012HB004) 云南省教育厅科研基金重点项目(2011Z015)
【分类号】：TP393.09

【相似文献】