Study of Gender-based Playing Style Stereotype in Overwatch using Machine Learning Analysis

Abstract

Since its release in 2016, Overwatch has become a game with a high amount of daily users (over 50 million users Worldwide) and made a great amount of profit for the company that produced it - Blizzard. However, it is common that the stereotypes and biases hide under the popularity of the games. In past researches, it is shown that gender could affect the designs, abilities, and player's view of in-game characters. In this research, a survey regarding gender stereotypes in Overwatch is conducted, and statistical and machine learning methods are applied to analyze the results. The result suggests that many of the players in the game had given biased or stereotypical actions or thoughts toward female players. Simultaneously, we propose an Ensemble Late Fusion (ELF) method that unifies the weighted predicted probabilities generated from various state-of-art classifiers to classify the player behaviors with or without gender-based stereotypes in Overwatch. The experimental results show that when using ELF, the optimal classification result reaches AUC to 0.9044, which beats several well-known traditional machine learning and deep learning classifiers. Moreover, the connection between certain biased ideas and rather a user in sexist is graphed and discussed in the paper.