Bridging Latent Factors and Tags: Enhancing Recommendation Systems
Vishal Paul · 2024
This paper explores the integration of latent factor models with item tags in recommendation systems to improve interpretability and effectiveness. Traditional latent factor models lack transparency, complicating understanding and trust. Leveraging a dataset from the Steam platform, I conduct exploratory data analysis and propose a novel model, tagMF, which generates latent factors based on tag relevance. The methodology includes model construction, data preprocessing, and evaluation based on mean squared error. Results show tagMF outperforming conventional models, enhancing interpretability, addressing the cold start problem, and offering personalized recommendations. Future enhancements include refining tag relevance and adopting Bayesian Personalized Ranking. This study underscores the importance of innovation in recommendation systems to meet evolving user needs.