A Two-Dimensional Affective Approach for Dynamic Difficulty Adjustment Based on Player Chat Analysis

Recent advances in generative artificial intelligence (AI) can transform video games by improving player–non-player character (NPC) interactions, which enable analysis of player language data and support dynamic difficulty adjustment (DDA) to enhance engagement. Understanding players’ perceptions of game difficulty is fundamental for DDA. Current assessment methods, however, suffer from high costs or genre specificity. Building on our previous research identifying hope and frustration as key difficulty indicators, we expanded our approach to include anxiety and boredom. We fine-tuned language models on annotated Dota 2 chat datasets, achieving over 97% accuracy. To ensure precision, our algorithm recommends increasing difficulty only when players express hope and boredom, and decreasing it when frustration and anxiety coincide. This strategy minimizes false positives in difficulty adjustment. Our method enables developers to personalize gameplay using chat-based affective analysis, providing a cost-effective, genre-agnostic way to assess players’ perceived difficulty and support a more comprehensive measurement approach.