Vector Embedding Techniques for Player Behaviour in DOTA 2

Personalization applications need prior information about a user, and popular techniques that use neural networks need data in a numerical format. This work explores three node embedding techniques to represent player behaviour in a compact vector form. I used data from the game DOTA 2 to produce vectors using three graph embedding methods, developed a testing framework, and conducted 270 experiments to explore the effect of different parameters on vector quality. I explored different values for parameters, including the number of vector dimensions, the number of games, the types of interactions used in training, and step size for updating vectors over time. The results show that using node2vec and a vector dimension of 512 outperforms the other methods in 13 of the 15 parameter variations that I tested. node2vec also computed the vectors 60 and 10 times faster than LINE and TGN, respectively, on the same amount of data.

Name thesis:degree_name

Master of Information Technology (M.I.T.)

Level thesis:degree_level

Master's

Discipline thesis:degree_discipline

Digital Media

Grantor dc:publisher

Carleton University

Year dc:date.issued

2021

Author dc:creator

• Shah, Jinay Gautamkumar

Statement dc:rights

• Copyright © 2021 the author(s). Theses may be used for non-commercial research, educational, or related academic purposes only. Such uses include personal study, research, scholarship, and teaching. Theses may only be shared by linking to Carleton University Institutional Repository and no part may be used without proper attribution to the author. No part may be used for commercial purposes directly or indirectly via a for-profit platform; no adaptation or derivative works are permitted without consent from the copyright owner.

Language dc:language.iso

en

DOI

10.22215/etd/2021-14680

OAI identifier oai:identifier

oai:carleton.scholaris.ca:20.500.14718/42416