基于KMeans-LR的广告推荐应用  

Advertising Recommendation Application Based on KMeans-LR

作　　者：郭尚志 廖晓峰[1] 赵庆 唐玉玲 GUO Shang-zhi;LIAO Xiao-feng;ZHAO Qing;TANG Yu-ling(College of Computer Science,Chongqing University,Chongqing 400030,China;Hunan Creator Information Technologies CO.,LTD.,Changsha Hunan 410205,China)

机构地区：[1]重庆大学计算机学院,重庆400030 [2]湖南科创信息技术股份有限公司,湖南长沙410205

出　　处：《计算机仿真》2024年第11期501-504,522,共5页Computer Simulation

摘　　要：随着互联网在线广告业务的飞速发展,面对海量稀疏的广告点击日志数据,单一传统型预测方法由于性能较低,在海量广告数据推荐上表现不佳。这里提出一种基于聚类-逻辑回归(KMeans-LR)的广告推荐模型,该模型首先通过Softmax函数对初始数据各列进行概率归一化,构造处理各列值为分类概率值;然后输入至KMeans算法,通过超参控制分类数,进行自动归类;最终输入分段逻辑回归模型进行预测。在公开数据集上进行实验,相对于传统的FM与逻辑回归算法,有更好的性能表现。With the rapid development of the Internet and online advertising business,in the face of massive and sparse advertising click log data,the single traditional prediction method has poor performance in massive advertising data recommendation due to its low performance.This paper proposes an advertising recommendation model based on cluster-logic regression(KMeans-LR).The model first normalizes the probability of each column of the initial data through the Softmax function,and constructs and processes each column value as the classification probability value.Then it is input into KMeans algorithm,and control the number of categories through hyper-parameters for automatic classification;Finally,it is input into the segmented logistic regression model for prediction.Compared with traditional FM and logistic regression algorithms,the experiment on open data sets has better performance.

关 键 词：多维聚类 逻辑回归 特征处理 智能推荐 人工智能与智能制造 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]