Verification of Dynamic Advertising Traffic Allocation Strategy Based on Multi arm Slot Machine

Qiquan Fang; Ziyi Liu; Zihan Meng; Jiarong Luo

doi:10.63313/EBM.9053

Authors

Qiquan Fang Zhejiang University of Science and Technology, Hangzhou 310023, China Author
Ziyi Liu Zhejiang University of Science and Technology, Hangzhou 310023, China Author
Zihan Meng Zhejiang University of Science and Technology, Hangzhou 310023, China Author
Jiarong Luo Zhejiang University of Science and Technology, Hangzhou 310023, China Author

DOI:

https://doi.org/10.63313/EBM.9053

Keywords:

Dynamic Advertising Traffic Allocation, Multi Arm Slot Machine, Thompson Sam-Pling Algorithm, Time Decay Factor

Abstract

With the rapid development of the Internet advertising market, advertisers have dramatically increased their demands for real-time accuracy. Traditional static allocation strategies have slow feedback and low exploration efficiency, while existing dynamic solutions still face problems such as complex calcula-tions and long cold start times. Therefore, this article proposes a lightweight dynamic allocation strategy based on the multi arm slot machine (MAB) framework. This strategy transforms advertising selection modeling into a seri-alized decision-making problem, providing real-time feedback to optimize long-term returns. The Thompson Sampling algorithm is used to introduce a time decay factor to address user interest drift. The experiment used Criteo's real advertising exposure datasets to compare five multi arm slot machine algo-rithms. The results showed that the Thompson Sampling algorithm with attenu-ation factor (TS Decay) performed excellently in terms of cumulative click through rate, convergence steps, and handling interest drift. In real data verifi-cation, the click through rate (CTR) reached 6.9%, which is better than other algorithms. This strategy combines Bayesian inference and time decay models to provide a framework for real-time decision-making in lightweight scenarios, helping to optimize the real-time performance of advertising delivery systems.

References

[1] Huang Xiaohui Research on User Experience Optimization and Conversion Enhancement Strategies for E-commerce Websites Based on User Behavior Analysis [J]. Mall Moderniza-tion, 2025, (10): 5-8

[2] Chen Jing Research on Sampling Techniques for Sparse Data in Recommenda-tion Systems [D]. University of Electronic Science and Technology of China, 2023. DOI: 10.27005/d.cnki.gdzku.2023.005854

[3] Yan Yang, Chen Zeqiu, Deng Junlin Heterogeneous Network Representation Learning Method Based on Multi arm Slot Machine [J]. Journal of Tianjin Vo-cational and Technical Normal University, 2024, 34 (01): 61-65

[4] Wu Bo, Fang Xifeng, Zhang Haixiang, et al. Research on dynamic programming methods for complex product design tasks combining fuzzy set theory and greedy algorithms [J]. Man-ufacturing Technology and Machine Tools, 2025, (03): 92-98. DOI: 10.19287/j.mtmt.1005-2402.2025.03.012.

[5] Xu Huanxuan. Research on Slot Machine Learning Algorithm Based on Thompson Sampling and Offline Data[D]. Chongqing University, 2023. DOI: 10.27670/d.cnki.gcqdu.2023.002777.

[6] Zhao Jieting. Construction of Recommendation Algorithm Integrating User Interests and Multi-Armed Bandit[D]. Shanghai International Studies Univer-sity, 2024. DOI:10.27316/d.cnki.gswyu.2024.000506.

[7] Wang Ding. Research on Recommendation Algorithms Based on Multi-Armed Bandit Mod-els [D]. Shanxi University, 2022. DOI: 10.27284/d.cnki.gsxiu.2022.002281.

[8] Zhang Fuchun. Research on Federated Learning Client Selection Mechanism Based on Combinatorial Multi-Armed Bandit[D]. Beijing University of Posts and Telecommunica-tions, 2022. DOI: 10.26969/d.cnki.gbydu.2022.001351.

[9] Ni Haojie. Research on the Impact of Information Flow Advertising Pushes on User Expe-rience under Algorithmic Technology [D]. Shanghai University of Finance and Economics, 2023. DOI: 10.27296/d.cnki.gshcu.2023.001659.

[10] Lu Zewei. Design and Implementation of a Personalized Information Recom-mendation System [D]. Beijing Jiaotong University, 2020. DOI: 10.26944/d.cnki.gbfju.2020.003546.