Email the Author
You can use this page to email Rafet Sifa about Matrix and Tensor Factorization for Profiling Player Behavior.
About the Book
Matrix and tensor factorization models are scalable and flexible tools for learning efficient representations for a variety of descriptive, predictive and prescriptive analytics tasks. They can be easily deployed without requiring high-performance computing environments and extensive parameter tuning for analyzing large scale and high dimensional behavioral data to come up with interpretable and actionable results. Their range of applications includes recommender systems, behavior prediction, natural language processing, digital forensics, process and budget optimization and behavioral profiling.
In this compact book, we will introduce certain theoretical as well as practical aspects behind a set of matrix and tensor factorization models for behavioral profiling. We will particularly concentrate on the various straightforward-to-implement algorithms used to come up with useful factorizations, ideas to enforce interpretability for human analysts and example behavioral profiling case studies with a behavioral dataset from a digital game.
The methods we cover in detail for profiling behavior include unconstrained matrix factorization, Singular Value Decomposition, k-Means clustering, Archetypal Analysis, Simplex Volume Maximization and k-Maxoids Analysis for analyzing bipartite (rectangular) matrices and two- and three-way DEDICOM for respectively decomposing structurally constrained asymmetric similarity matrices and tensors.
Although our main intention is about analyzing player behavior in digital games, the methods we introduced can easily be applied to analyze behavioral telemetry data from similar digital products (such as mobile and web applications) as well. Therefore, this book is suited for developers, analysts, data scientists and engineers, who are interested in tracking, collecting and analyzing behavioral telemetry data to better understand their user-base and automatically learn informative features for a variety of analytics applications.
This is the alpha version of the book containing all the methods we intend to cover and the final edition is expected to be released in February 2019.
About the Author
I am the deputy head of the Media Engineering Department, head of the business unit Cognitive Business Optimization and a senior research data scientist at Fraunhofer Institute for Intelligent Analysis and Information Systems (IAIS) in Sankt Augustin Germany. I am also currently teaching Msc. level data science courses as a guest lecturer to students of University of Bonn and RWTH Aachen University at Bonn-Aachen International Center for Information Technology. My current research focus is based on statistical data mining in the context of representation learning for a variety of industry applications involving behavioral analytics, digital forensics and text mining. Before joining Fraunhofer, I worked as a data scientist in the games industry, where I mainly concentrated on developing large scale machine learning solutions for business intelligence problems. I often give talks and organize workshops about representation learning and behavioral analytics in scientific conferences and industrial summits.