Benchmarking CTR Prediction Models in Sponsored Search: A Comparative Study

In this paper, we explore a new approach to predicting click-through rates (CTR) in online advertising called Factorization Machines (FMs). CTR is an important metric for measuring the success of ad campaigns, but it can be challenging to predict accurately. Traditional methods rely on manually crafted combinatorial features, which can be time-consuming and difficult to maintain.
Our proposed approach uses a novel combination of factorization machines and neural networks to learn high-level representations of ads and users. FMs are a type of non-negative matrix factorization technique that can identify latent factors in large datasets. By combining FMs with neural networks, we can learn complex interactions between ad features and user contexts, leading to improved CTR predictions.
To evaluate our approach, we conduct a comprehensive evaluation on several benchmark datasets from different domains. Our results show that FMs outperform state-of-the-art baselines in terms of CTR prediction accuracy. We also observe that the learned factorization matrices have interesting structures and can be used to identify meaningful patterns in the data.
In addition, we conduct a series of ablation studies to analyze the effectiveness of different components in our approach. These studies show that the combination of FMs and neural networks is essential for achieving good performance, and that incorporating user contexts improves predictions further.
Our work has important implications for the field of recommender systems and computational advertising. By developing a more accurate and efficient CTR prediction method, we can help advertisers optimize their campaigns and improve their return on investment. Moreover, our approach is interpretable and easy to implement, making it accessible to practitioners and researchers alike.
In summary, Factorization Machines offer a powerful new tool for predicting click-through rates in online advertising. By combining the strengths of factorization machines and neural networks, we can learn complex patterns in ad data and improve CTR predictions with superior accuracy. Our approach has important implications for optimizing ad campaigns and improving the return on investment for advertisers.