Product sales recommendation system using item-based collaborative and content-based filtering

In the fast-evolving world of e-commerce, improving user experience and operational efficiency emerges from the utilization of recommendation systems. Unlike other systems that mostly center on providing product suggestions to consumers, this study proposes a new system for sellers which assists them with insights on how to better their sales strategies. The system proposed in this study implements a Content-Based Filtering with Facebook AI Similarity Search (FAISS) algorithm and Collaborative Filtering with Neural Collaborative Filtering (NCF). To allow effective product similarities by FAISS, text data is first processed with the Sentence-BERT model that generates semantic embeddings. Also, NCF customizes the predictive outcomes using user-item interaction patterns. The hybrid approach overcomes challenges such as the cold start and data sparsity issues, typical in system initialization phases. Through experiments, the hybrid model performed significantly better than the pure models. A dataset of over one million entries yielded a Hit Rate of 84.91%, Precision of 51.10%, and Recall of 83.38%. These results demonstrate the advantage of hybrid filtering approaches, particularly in large-scale, diverse e-commerce data, to provide precise and easily scaled product recommendations.