Advertising Wisely: A Comprehensive Study of State-of-the-Art Recommendation Models for Cold-Start Scenarios in Online Magazine Advertising

Recommender systems are crucial for traffic driven online media platforms, aligning relevant advertisements with target audiences to enhance engagement and revenue. However, selecting effective methods remains challenging, especially under cold-start conditions where user interaction data is sparse. This study evaluates 13 state-of-the-art recommendation models—including rule-based, machine learning, and deep learning methods—using eight standard metrics, providing actionable insights for online advertising practitioners. Additionally, we propose as impleyet effective strategy to mitigate the cold-start issue by reformulating interaction data to increase its density. Our results show that: (1) deep learning models typically outperform classical machine learning approaches in cold-start scenarios; (2) the proposed data reformulation significantly improves accuracy across all tested models without reducing recommendation diversity; and (3) when data sparsity decreases, light weight machine learning methods can outperform complex deep learning models, offering practical and efficient solutions for real-world deployment.

Keywords: Online Advertising, Recommendation System, Cold-Start Problem, User Interaction Data, Machine Learning.