Experience is all you need: A large language model application of fine-tuned GPT-3.5 and RoBERTa for aspect-based sentiment analysis of college football stadium reviews

What really moves the needle in stadium satisfaction isn’t just more points on the board. It’s fewer pain points in the journey. Our study of 8,405 TripAdvisor reviews on Power Five college football venues shows a clear pattern: elevate the highs (core game thrills, smooth operations, electric atmosphere, shared traditions) and, more importantly, aggressively eliminate the lows, especially functional hassles and safety concerns. Even small reductions in negative functional and safety experiences translate into meaningful jumps in the odds of a five-star rating. In other words, delight matters, but friction costs more.

What I love here is the blend of rigor and relevance. The authors build a unified customer experience lens (core, functional, emotional, socialization, monetary, safety) and then operationalize it with an LLM pipeline: one fine-tuned GPT-3.5 to extract aspects, another to classify experiences, and a fine-tuned RoBERTa to score sentiment. That post-positivist, “qual + quant” approach turns messy fan narratives into decision-ready signals for game-day design and investment.

For practitioners: double down on emotion and community, but set a zero-tolerance policy for functional friction and perceived insecurity. For scholars: this is a replicable, sport-specific NLP workflow that travels well to other venues, leagues, and platforms.