Combining XGBoost and hybrid filtering algorithm in e-commerce recommendation system

Abstract

This study proposes a hybrid filtering algorithm (HFA) that combines extreme gradient boosting (XGBoost), content-based filtering (CBF), and collaborative filtering (CF) to improve recommendation accuracy in electronic commerce (e-commerce). XGBoost first leverages demographic data (e.g., age, gender, and location) to address cold start conditions, producing an initial product prediction; CBF refines this prediction by measuring product similarities through term frequency-inverse document frequency (TF-IDF) and cosine similarity, while CF (implemented via singular value decomposition) further incorporates user interaction patterns to enhance recommendations. Experimental results across multiple datasets demonstrate that HFA consistently outperforms standalone XGBoost in key metrics, including precision, F1-score, and hit ratio (HR). HFA’s precision often exceeds 90%, indicating fewer irrelevant recommendations. Although recall levels remain modest, HFA exhibits stronger adaptability under cold start scenarios due to its reliance on demographic features and user-item interactions. These findings highlight the efficacy of combining advanced machine learning with hybrid filtering techniques, offering a more robust and context-aware solution for e-commerce recommendation systems.