Learning about media Users from movie rating data

Abstract

In this thesis, we study techniques of machine learning for media users who submitted movie ratings to the MovieLens dataset --- a project inspired by Sky UK's own business problems I encountered during my internship there. It follows the "feature engineering" paradigm, compared to the "deep learning" paradigm, through three stages: Feature Engineering, Clustering and Recommendation, each being a classic machine learning problem. For each step, I am introducing the common, relevant methods, along with my own designed models on top of available tools and experiments on the MovieLens data on the Google Cloud Platform. Due to the open-ended nature of all three problems, we don't have quantifiable conclusions on which methods would prove the best; instead, presented here is some learning on the trade-offs and suitability for these designs.