DEVELOPING A NEXT GENERATION RECOMMENDER SYSTEM TO ENHANCE USER EXPERIENCE

Abstract

For business expansion, recommender systems have been shown to be the most crucial technology. Recommender systems employ intelligent algorithms to ascertain a user's preferences and offer pertinent recommendations. This thesis explores an innovative approach to movie recommendation by using BERT (Bidirectional Encoder Representations from Transformers) to construct embeddings for movie titles from the MovieLens dataset. These embeddings are used to produce the cosine similarity scores that form the basis for movie recommendations. The steps in the procedure are extracting embeddings, comparing them with collaborative filtering based on fine tuning a pretrained BERT model. The accuracy, mean squared error (MSE), and a few other measures are used to evaluate how effective the BERT-based recommendations are. The findings show potential in terms of determining semantic relationships between film titles. This work highlights the potential of combining advanced natural language processing methods with traditional recommendation systems to improve the accuracy and relevance of cinematic recommendations and provide important insights into personalized content delivery. This study also demonstrates the scalability of using BERT embeddings for huge datasets, providing a solid foundation for further recommendation system research. By using cutting-edge NLP techniques, new opportunities for enhancing user happiness through more precise and tailored recommendations are created. This method greatly advances the field of content recommendation technology while also optimizing computational resources.