Integrating Explicit and Implicit Feature Interactions for Online Ad Installation Forecasting

We present our solution for the RecSys Challenge 2023 in this paper, which focuses on online advertising and deep funnel optimization, emphasizing user privacy. The dataset provided for the challenge includes user and ad features, as well as click and install information from the ShareChat apps. The objective is to predict the probabilities of ad installations in the test set. Our solution primarily leverages the xDeepFM model, which combines explicit and implicit feature interactions to capture complex relationships. Additionally, we employ various techniques such as feature engineering, feature crossing, cross-validation, and model integration to enhance the performance of our solution. Through extensive experimentation and fine-Tuning, our team BUAA_BIGSCity achieved a score of 6.282142 in the final submission, demonstrating the effectiveness of our approach. To promote reproducibility and further research, our code is available on GitHub 1. This paper provides insights into our solution for this challenge, showcasing advancements in online advertising and deep funnel optimization.