Modeling Individual Differences in Game Behavior Using HMM

Abstract

Player modeling is an important concept that has gained much attention in game research due to its utility in developing adaptive techniques to target better designs for engagement and retention. Previous work has explored modeling individual differences using machine learning algorithms performed on aggregated game actions. However, players’ individual differences may be better manifested through sequential patterns of the in-game player’s actions. While few works have explored sequential analysis of player data, none have explored the use of Hidden Markov Models (HMM) to model individual differences, which is the topic of this paper. In particular, we developed a modeling approach using data collected from players playing a Role-Playing Game (RPG). Our proposed approach is two fold: 1. We present a Hidden Markov Model (HMM) of player in-game behaviors to model individual differences, and 2. using the output of the HMM, we generate behavioral features used to classify real world players’ characteristics, including game expertise and the big five personality traits. Our results show predictive power for some of personality traits, such as game expertise and conscientiousness, but the most influential factor was game expertise.

Predicting Risk Propensity Through Player Behavior in DOTA 2: A Cross-Sectional Study

“…It possibly signified that GPR was a relatively suitable algorithm for DOTA 2 behavioral datasets. Some researchers have investigated how to predict players' psychological traits through in-game behavior, but the results did not show strong predictive power (Yee et al, 2011a;Fong and Mar, 2015;Bunian et al, 2017;Wu et al, 2021). For example, Bunian et al (2017) used hidden Markov models (HMM) to extract the sequence of players' actions and trained classification models for the Big Five personality traits.…”

“…For multiplayer online battle arena games and first-person shooter games, researchers reported how the role preference and game actions entangled with players' Big Five personalities (Wang et al, 2019) and other personality traits such as aggression (Delhove and Greitemeyer, 2020). In terms of personality prediction, a number of studies used in-game behaviors to realize personality classification and regression (Bunian et al, 2017;Ammannato and Chiesi, 2020). These predictive models provide new approaches for perceiving individuals' psychological characteristics in a nonintrusive way while traditional measurements are not applicable.…”

Attribute Inference Attacks in Online Multiplayer Video Games: a Case Study on Dota2

“…We grouped age in three bins (similarly to [14]): very young/underage, young adults, and over 25 (our 'oldest' respondent was 38); the frequencies for purchase_habits are never, less than once a month (rarely), and monthly or more often (regularly); for occupation, we consider a student as unemployed. Since our survey quantified each personality trait as an integer [0-100], we group such values into three categories (similarly to [11]) differentiating low, middle, or high scores.…”

“…Overall, we compute over 300 features, 11 which identify three datasets: P, focused on the players, containing 484 samples, each described by 187 features; M, focused on the matches, containing 26241 samples, each described by 137 features; and M, containing 11117 samples and 160 features, which is a 'distilled' version of M. In particular, M differs from M in two ways: First, we address the problem of the highly imbalanced distribution of M in terms matches-per-player (some players in A have only 5 matches in M, while others have hundreds); we thus reduce the potential bias by randomly sampling 12 at most 30 matches for each player. Second, we augment the features in M with those derived with our domain knowledge; the intention is determining how much of an impact our intuitions have on all our experiments.…”

Understanding human decision-making in supply chains: using serious gaming for modeling action, thought, and the environment

“…Other researchers used HMMs for modeling player actions in the game [150,151]. HMMs have shown great promise, especially for analyzing sequential data [152] and modeling cognitive processes [153].…”