Improving Recommender Engines by Integrating Spatial Statistics
Date
2020
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The web has become a significant medium for business transactions and e-commerce. With such a vast quantity of options for users to choose and buy, recommender engines have been created to analyze patterns of user interests in products. These engines help to tailor recommendations that users are likely to enjoy buying based on their previous purchases or explicit feedback on their likes and dislikes. There are numerous algorithms to build recommender engines but the one that became most popular was Simon Funk's Singular Value Decomposition (SVD) for the Netflix Challenge. He demonstrated that matrix factorization models are superior to classic nearest-neighbor collaborative filtering techniques for producing product recommendations [3]. The main scope of this dissertation is to develop effective and efficient techniques that improve upon Simon Funk's SVD by thoughtfully integrating spatial statistics in such a way that will significantly improve the prediction error. Based on the type of data (whether it may be more spatially autocorrelated or latent), the method to blend both algorithms together can be quite different. The contribution of this dissertation is focused on building unique models that integrate SVD and kriging in different ways based on the type of data given to understand when to use them and why.