Recognizing Affiliation: Using Behavioural Traces to Predict the Quality of Social Interactions in Online Games
Authors:
Julian Frommel,
Valentin Sagl,
Ansgar E. Depping,
Colby Johanson,
Matthew K. Miller,
Regan L. Mandryk
Abstract:
Online social interactions in multiplayer games can be supportive and positive or toxic and harmful; however, few methods can easily assess interpersonal interaction quality in games. We use behavioural traces to predict affiliation between dyadic strangers, facilitated through their social interactions in an online gaming setting. We collected audio, video, in-game, and self-report data from 23 d…
▽ More
Online social interactions in multiplayer games can be supportive and positive or toxic and harmful; however, few methods can easily assess interpersonal interaction quality in games. We use behavioural traces to predict affiliation between dyadic strangers, facilitated through their social interactions in an online gaming setting. We collected audio, video, in-game, and self-report data from 23 dyads, extracted 75 features, trained Random Forest and Support Vector Machine models, and evaluated their performance predicting binary (high/low) as well as continuous affiliation toward a partner. The models can predict both binary and continuous affiliation with up to 79.1% accuracy (F1) and 20.1% explained variance (R2) on unseen data, with features based on verbal communication demonstrating the highest potential. Our findings can inform the design of multiplayer games and game communities, and guide the development of systems for matchmaking and mitigating toxic behaviour in online games.
△ Less
Submitted 6 March, 2020;
originally announced March 2020.
