Research Information System
33rd IEEE Conference on Signal Processing and Communications Applications, SIU 2025, İstanbul, Turkey, 25 - 28 June 2025, (Full Text)
Cross-domain recommender systems aim to im- prove recommendations in the target domain by exploiting the source domain where abundant data is present. In this study, we investigate the combination of a shallow model, the factorization machine, with cross-domain models to improve click-through rate prediction, which has a significant impact on online advertising systems. Deep neural network based models perform well in predicting click-through rates and are strong at capturing non- linear interactions. However, while deep models capture complex correlations, they may miss superficial relationships between users and items. In this study, the integration of a shallow model, the factorization machine, with deep cross-domain models is presented as an important approach to capture low-order interactions and obtain more meaningful embedding vectors. Experiments show that the combination of FM and cross-domain models leads to a significant performance improvement in click-through rate prediction.