基于用户点击的线性回归在内容推荐中的应用研究

2017-09-08 00:00:56石方夏
现代电子技术 2017年17期

石方夏

摘 要: 在内容推荐中根据用户的浏览偏好进行内容排序对提高用户的点击率具有至关重要的作用。推荐流中内容随着时间变化呈现出流动性,分析历史数据中用户和推荐内容的点击信息进行回归分析,提取用户在内容点击时特征的相关性,并对特征进行归一化,拟合出在当前特征分布下的点击率,以线性回归作为拟合模型进行用户点击率预测。实验以今日头条中用户浏览点击日志为测试数据集,采用改进算法进行内容排序时截取单领域下用户点击内容作为点击率进行验证,实验结果表明,改进算法能够较为准确地推荐用户倾向点击的内容。

关键词: 用户点击率; 线性回归; AUC; 特征抽取; 回归分析

中图分类号: TN911?34; TP391 文献标识码: A 文章编号: 1004?373X(2017)17?0135?03

Application of linear regression based on user clicks rate in content recommendation

SHI Fangxia

(School of Information Engineering, Xizang Minzu University, Xianyang 712082, China)

Abstract: The content ranking according to users′ browsing preference in content recommendation plays an important role in improvement of the user clicks rate. The content in recommendation flow changes with time. The clicks information of user and recommendation content in historical data is analyzed for regression analysis. The feature correlation while user clicks the content is extracted. The features are normalized to fit the clicks rate of current features distribution. The linear regression is used as the fitting model to predict user clicks rate. The logs browsed and clicked by users are taken as the test dataset in the experiment. The content clicked by users in the single field is cut out with the improved algorithm as the clicks rate for verification. The experimental results show that the improved algorithm can recommend clicks content of user preference accurately.

Keywords: user clicks rate; linear regression; AUC; feature extraction; regression analysis

0 引 言

目前,國内新闻类产品如今日头条、一点资讯和微博头条等都以内容推荐为主,如何提高内容推荐的准确性,从而提升用户对推荐内容的点击率,提升用户对平台的粘性,对提升平台的用户留存具有非常重要的意义。

用户在内容浏览时的用户行为至关重要,用户对内容的点击、评论和转发等行为都能很明显地反映用户对内容的偏好程度,分析抽取用户在内容浏览时的反馈行为能够很好地帮助提升平台内容排序的准确性。本文尝试以用户在内容浏览时对各个领域点击分布为出发点,抽象出用户对各个领域的兴趣度,进而分析提取用户对各个领域的点击倾向性,并结合内容的时间因素和热度因素进行综合回归分析,基于线性回归中各个特征的强解释性,采用线性回归进行各个特征的回归系数拟合。……

登录APP查看全文