Recommendation System
#1. Bayesian Approach to Movie Recommendation Model
I am excited to showcase my work on the enhancement and refinement of the Variational Probabilistic Matrix Factorization Algorithm, as outlined in the seminal paper titled "Variational Bayesian approach to Movie Rating Prediction." This project involved a comprehensive analysis and restructuring of the algorithm, with a particular focus on the incorporation of Variational Expectation-Maximization techniques.
Leveraging Mean Field Approximation for the E-step, I meticulously trained the model on a substantial dataset containing 100,000 observed values within the matrix. One of the key achievements of this project was the assessment of Root Mean Square Error (RMSE) discrepancies between a baseline initialization and an alternative approach for the decomposed matrices U and V. To enhance model convergence, I devised a strategy that harnessed values from the initial iteration of EM-SVD, resulting in a remarkable 10% increase in convergence rates. This project represents my dedication to pushing the boundaries of machine learning algorithms and achieving tangible improvements in predictive accuracy and model performance.