陈彦彬1,2  杨泽华2  谢佳2

（1.揭阳职业技术学院实训与信息中心，广东 揭阳 522051

2.广东博华科技有限公司，广东 揭阳 522000）

摘要：针对电梯传媒终端广告精准投放面临的采集广告受众个人隐私数据难的问题，提出融合电梯交通流量、广告主行为时空特征、广告主题特征、广告主评分行为等多源特征的电梯广告推荐算法。首先，利用差分函数算法从电梯运行数据中提取电梯交通流量峰值特征；然后，通过广告主广告行为数据提取广告主行为时空特征；接着，将电梯交通流量峰值特征与广告主行为时空特征融合，并利用ReliefF算法进行特征筛选；最后，设计融合多源特征的电梯广告推荐系统，实现电梯广告节目的精准投放。实验结果表明：融合多源特征的电梯广告推荐算法的Precision、Recall和ROC曲线的AUC值等评价指标均明显提高；在一定程度上解决了冷启动、数据稀疏等问题。该系统无需采集广告受众的个人隐私数据，具有较强的实用性。

关键词：多源特征；电梯广告；推荐算法；精准投放

中图分类号：TP 301             文献标志码：A          文章编号：1674-2605(2023)02-0006-09

DOI：10.3969/j.issn.1674-2605.2023.02.006

Elevator Advertising Recommendation System Integrating              Multi-source Features

CHEN Yanbin1, 2  YANG Zehua2  XIE Jia2

(1. Training and Information Center, Jieyang Polytechnic, Jieyang 522051, China

2. General Manager, Guangdong Bohua Technology Co., Ltd., Jieyang 522000, China)

Abstract: In response to the difficulty in collecting personal privacy data of advertising audiences for precise advertising placement in elevator media terminals, a elevator advertising recommendation algorithm is proposed that integrates multi-source features such as elevator traffic flow, spatiotemporal characteristics of advertiser behavior, advertising theme characteristics, and advertiser rating behavior. Firstly, the difference function algorithm is used to extract the peak characteristics of elevator traffic flow from elevator operation data; Then, extract the spatiotemporal characteristics of advertisers' behavior through their advertising behavior data; Next, the peak characteristics of elevator traffic flow are fused with the spatiotemporal characteristics of advertiser behavior, and the ReliefF algorithm is used for feature selection; Finally, design an elevator advertising recommendation system that integrates multi-source features to achieve accurate placement of elevator advertising programs. The experimental results show that the evaluation indicators such as Precision, Recall, and AUC value of the ROC curve of the elevator advertising recommendation algorithm that integrates multi-source features are significantly improved; To some extent, it has solved problems such as cold start and data sparsity. This system does not need to collect personal privacy data of advertising audiences, and has strong practicality.

Keywords: multi-source features; elevator advertising; recommendation algorithm; accurate placement