Advancing E-commerce Recommendations: A Comparative Study of BERT-Enhanced Collaborative Filtering and Alternative Approaches

In e-commerce, recommendation systems are critical to increasing user engagement and conversion. However, traditional collaborative filtering methods require user feedback and do not take into consideration semantic understand of the items. In order to eliminate these barriers, this work suggests a novel solution of fusing collaboration technique with BERT-based embeddings for e-commerce suggestions. Besides this we include a variety of other models for comparison namely standalone BERT, Collaborative filtering approach, Roberta + Collaborative Filtering and Deberta+Collaborative Filtering. We are using collaborative filtering and item title, reviews (via BERT-based embeddings) to model user preferences. Through experiments on a real Flipkart dataset, we validate the effectiveness of our approach. Recommendation in the merged model results, is our proposed model achieved higher accuracies. These results hold valuable insights for enhancing e-commerce platform recommendation systems. This study helps improve the state-of-the-art in e-commerce recommendation systems by combining collaborative filtering and more advanced natural language processing methods.