Mining Player Experience Trends From Game Reviews Using Large Language Models

How have player experiences changed over the years? For instance, have there been general shifts in what kinds of emotions players experience and express? We probe these questions with help of recent methodological advances in psychology and Large Language Models (LLMs), in particular the possibility to predict Likert-scale responses based on free-form text. Applying this at scale to three player experience questionnaires (PXI, CORGIS, AESTHEMOS) and 152143 Metacritic user reviews from years 2010-2024, we reveal trends such as an increasing portion of reviews expressing emotional challenge, meaning, and nostalgia. We then analyze the contributions of different genres and games to the trends, in addition to reasons explicitly indicated by the reviews, and establish correlations between review scores and different player experience constructs. Taken together, our results provide novel insights into how player experiences have evolved. Methodologically, we propose and demonstrate a novel and scalable method for analyzing game reviews.

• Human-centered computing → Empirical studies in HCI.