The world of online multiplayer gaming has reached a point where player communities have become plagued with verbal toxicity in the form of hate and harassment, a detrimental issue for both players and game development companies alike. This master's thesis involves the research, design and development of an intervention system that can be adapted and integrated into online games. The goals of the intervention system are to detect, moderate, and potentially prevent verbal toxicity. Design thinking was used as a methodology to promote player-centered interfaces and interventions. Deep learning-based natural language processing (NLP) techniques were used to develop the back-end toxicity detection. The NLP algorithms process speech to analyze two modalities of verbal toxicity: text transcription and audio features. Investigation of audio feature analysis was prioritized since it is less prevalent in the existing literature and intervention systems. Audio feature extraction was explored to identify an optimal set of features that are both measurable and indicative of toxicity. The intervention system responds to detected toxicity by overlaying interface elements during gameplay to moderate and prevent verbal toxicity while accommodating diverse scenarios and player needs. User experience heuristics in learning, feedback, visual appearance and interaction were applied alongside player testing with 10 participants to develop a refined prototype that prioritizes strong player experience, player-system cooperation, and operant conditioning to correct toxic behavior.

• Author