Author

Trevor Flint

Date of Award

2022

Document Type

Thesis

Degree Name

Bachelors

Department

Natural Sciences

First Advisor

Gillman, David

Area of Concentration

Computer Science

Abstract

The video game DotA 2 has a steep learning curve. Choosing which character (or “hero”) and position to play is not obvious for new players. This thesis proposes a recommender system that chooses a hero and position for a player based on their style of play. This system works by matching the player with experts who play similarly. The system is intended to make the game more enjoyable for beginners and to increase player retention. In this work, we present a dataset containing the actions of expert players representing different heroes in different positions and situations over many thousands of games. We present the results of clustering algorithms applied to heroes and positions according to their actions and applied to the actions themselves for feature selection. We present the results of an initial machine learning model that uses these features to predict hero and position from a player’s style of play.

Share

COinS