基于LightGBM的广告商品平台推荐系统设计与应用  

Design and Application of a Recommendation System for Advertising Commodity Platform Based on LightGBM

作　　者：杨正成 刘浩[1] YANG Zhengcheng;LIU Hao

机构地区：[1]东华大学信息科学与技术学院,上海201620

出　　处：《科技创新与应用》2022年第30期1-6,共6页Technology Innovation and Application

基　　金：新一代广告创意平台校企合作课题(20210107D)。

摘　　要：字节跳动旗下的广告创意定制平台,是为广告主提供广告商品创意定制服务的撮合平台,该文旨在为该平台搭建基于LightGBM算法的商品推荐系统,以提升平台商品服务的点击率(CTR)与下单转化率。该文对软件体系架构进行优化设计,包括构建个性化CTR预估模型、优化推荐排序策略,即采用LightGBM算法的精排模型对商品平台中的服务进行个性化排序推荐。经过实际的线上验证,设计的推荐系统能够帮助平台有效提升流量转化数据,其中广告商品CTR上涨10.52%,下单转化率增长79.3%。因此,该推荐系统可在商业应用中为广告主减少广告商品选购的决策成本,并为平台带来增量营收。The advertising commodity platform of ByteDance is a bilateral trading platform that provides advertising creative customization services for advertisers.This paper aims to build a product recommendation system based on the LightGBM algorithm to improve the click-through rate(CTR)and purchase conversion rate of products and services on the platform.Therefore,the overall software architecture is designed,which includes the construction of personalized CTR prediction model,and the optimization of the recommendation ranking strategy,where the fine line model of the LightGBM algorithm is adopted to implement the personalized ranking recommendation for the services of the commodity module.Through actual online verification,the optimized recommendation system helped the platform to effectively improve the business data.The CTR of advertising products increased by 10.52%,and the purchase conversion rate increased by 79.3%.In commercial applications,this designed system can reduce the decision-making cost of advertising products selection by advertisers and thus bring the incremental revenue to the platform.

关 键 词：广告服务推荐 点击率预测 LightGBM 下单流程 软件体系架构 

分 类 号：TP312[自动化与计算机技术—计算机软件与理论]