Javascript must be enabled to continue!
Explainable Visitor Profiling for Adaptive Cultural Heritage Experiences

Adaptive cultural heritage experiences require an understanding of why visitors engage, how they prefer to explore, and how much information they seek. We present an explainable visitor-profiling service to provide an initial motivation state for personalised cultural heritage experiences. The service models five profiles: StorySeeker, SocialVisitor, Fast-trackVisitor, In-depthExplorer, and Recharger.

The empirical dataset contains 189 questionnaire responses and 19 behavioural and preference-related features. Strong class imbalance, overlapping motivations, and the absence of observed In-depthExplorer respondents made the direct five-class training unreliable. We therefore generated 50,000 synthetic profiles from the empirical response distributions, supplemented with an expert-defined representation of the missing profile. Candidate classifiers and imbalance handling strategies were trained on the synthetic data and evaluated on the observed responses. XGBoost with SMOTE achieved the highest balanced accuracy, at 71.0%, across the four profiles represented in the observed data.

The service separates prediction from explanation. The classifier produces profile probabilities and an uncertainty indicator, while local SHAP values identify supporting and opposing questionnaire evidence. An open-weight multilingual language model converts this structured evidence into a concise explanation without changing the predicted profiles or probabilities. A containerised FastAPI implementation supports structured and streamed responses in eight tested languages.

Future work will integrate a Motivation State Agent that updates profile probabilities from visitor behaviour and feedback, enabling adaptive content, storytelling, recommendations, and clarification questions throughout the visit.

Attachment language: English File type: PDF document Presentation Poster
Updated: 29-09-2026 16:59 - Size: 1.02 MB
Attachment language: English File type: PDF document Presentation File
Updated: 31-10-2026 11:11 - Size: 458.92 KB
Speakers
Vasilis Perifanis

Back

ACM GreekCHI-CYCHI Symposium 2026


We are excited to invite you to the ACM GreekCHI-CYCHI Symposium 2026, held on November 6-7, 2026 jointly in Athens, Thessaloniki, Nicosia and Online.