The Effect of Context-aware LLM-based NPC Conversations on Player Engagement in Role-playing Video Games
Author
Csepregi, Lajos Matyas
Term
4. term
Education
Publication year
2023
Pages
19
Abstract
Denne afhandling undersøger, om samtaler med ikke-spillerkarakterer (NPC’er) i rollespil bliver mere engagerende, når de genereres af store sprogmodeller (LLM’er), der kan bruge kontekst. Med kontekst menes, at dialogen tager højde for, hvad der allerede er sagt, og hvad der sker i spillet, herunder spillerens tidligere valg og spillets aktuelle tilstand. Vi byggede NPC-samtaler med en LLM og testede dem i et eksperiment med 21 deltagere. Vi målte spillerengagement og deltagernes vurdering af samtalekvalitet. Resultaterne peger på, at kontekstsensitive, LLM-baserede NPC-dialoger har stort potentiale til at øge spillerengagement i RPG’er. På baggrund af studiet præsenterer vi også praktiske designretningslinjer for at implementere kontekstsensitive NPC-interaktioner. Afhandlingen anerkender studiets begrænsninger og anbefaler yderligere forskning for at belyse andre aspekter af emnet.
This thesis explores whether conversations with non-player characters (NPCs) in role-playing games become more engaging when they are generated by large language models (LLMs) that can use context. Here, context means the dialogue takes into account what has already been said and what is happening in the game, including the player’s prior choices and the current game state. We built NPC conversations using an LLM and tested them in an experiment with 21 participants. We measured player engagement and how participants rated the quality of the conversations. The results suggest that context-aware, LLM-based NPC dialogue has strong potential to increase player engagement in RPGs. Based on the study, we also outline practical design guidelines for implementing context-aware NPC interactions. The thesis acknowledges the study’s limitations and recommends further research to explore other aspects of the topic.
[This abstract was generated with the help of AI]
Keywords
Documents