A Multi-Objective Recommender System for Enhanced Consumer Behavior Prediction in E-Commerce

With the proliferation of online shopping, consumers are inundated with an overwhelming number of product choices, often leading to decision fatigue and abandoned shopping carts. To mitigate this issue, recommender systems have been employed to personalize the shopping experience and enhance consumer engagement. However, traditional recommender systems typically focus on single-objective optimization, such as predicting clicks or purchases, which limits their effectiveness. In this study, we propose a multi-objective recommender system designed to simultaneously optimize multiple consumer behaviors, including clicks, cart additions, and completed orders. Leveraging the OTTO Recommender Systems Dataset, our system employs a two-stage candidate reranker model that first generates a pool of potential items and then ranks them based on predicted relevance. Our experimental results demonstrate that an ensemble model combining LightGBM, XGBoost, and CatBoost Rankers outperforms individual models, achieving superior recall rates. This multi-objective approach provides a comprehensive solution for enhancing the accuracy and effectiveness of recommender systems in e-commerce.