How long visitors spend viewing artworks, often referred to as dwell time, has long been studied in museology as a potential behavioral indicator of engagement. However, dwell time may encode both genuine preference and situational factors such as fatigue, and disentangling these signals for computational modeling has received limited attention. This study investigates whether dwell time can serve as a valid preference indicator for implicit user modeling and whether incorporating it can improve artwork recommendation. Using the BIRD dataset, which includes eye-tracking data for extracting dwell times and explicit preferences from 51 museum visitors, we report three main findings. First, visitors spend significantly longer (9.27 seconds on average) viewing artworks they like, with a large effect size (Cohen’s d = 1.47). Second, we confirm the museum fatigue phenomenon, the progressive decline in visitor attention throughout a visit, observing a 34% reduction in dwell time from visit start to end. Third, we evaluate collaborative filtering approaches and find that while purely implicit models using dwell time alone perform near-chance level, a hybrid approach that uses dwell time to compute item similarities while predicting preferences from explicit likes achieves the best performance (AUC-ROC = 0.755, AP = 0.522). These findings suggest that dwell time provides complementary information to explicit feedback and can enhance museum recommendation systems when appropriately integrated.

Abstract Museums widely use audio guides, yet these are delivered identically to all visitors regardless of their profile. This study addresses that gap by combining GPT-4 with Falk and Dierking’s visitor categorization framework to generate personalized audio guides and examine whether LLM-generated content can effectively meet the distinct needs of five visitor types: Explorers, Experience Seekers, Professionals/Hobbyists, Facilitators, and Rechargers. Personalized and general audio guides were generated for three artworks using a prompt-chain approach encoding visitor-specific needs. A user study with 56 self-identified participants evaluated the content, complemented by expert validation assessing factual accuracy and domain-specific quality. Personalized guides outperformed general ones across most categories and needs, with notable gains in engagement for Experience Seekers (+8.9%), curiosity stimulation for Explorers (+8.8%), and work/hobby relevance for Professionals/Hobbyists (+10.1%). Preference, however, was moderated by artwork characteristics. Expert review confirmed factual soundness while identifying gaps in domain-specific terminology and tonal alignment. These findings directly motivated our Curator-AI collaborative framework, which organizes content production into four phases, with the curator as the governing intelligence. This work demonstrates the potential of LLMs to support personalization in cultural heritage contexts through established visitor categorization frameworks. It also highlights the importance of a human-in-the-loop approach, in which curators remain actively involved in supervising, validating, and refining AI-generated content. Overall, the proposed framework suggests a viable path toward scalable and personalized museum experiences that balance automation with curatorial oversight.

Indoor Positioning Systems (IPSs) hold significant potential for enhancing visitor experiences in cultural heritage. By enabling personalized navigation, efficient artifacts organization, and better interaction with exhibits, IPSs can transform how individuals engage with museums, galleries and libraries. However, these institutions face several challenges in implementing IPSs, including environmental constraints, technical limits, and limited experimentation. Received Signal Strength (RSS)-based approaches using Bluetooth Low Energy (BLE) and WiFi have emerged as preferred solutions due to their non-invasive nature and minimal infrastructure requirements. Nevertheless, the lack of publicly available RSS datasets that specifically reflect museum environments presents a substantial barrier to developing and evaluating positioning algorithms designed for the intricate spatial characteristics typical of cultural heritage sites. To address this limitation, we present BAR, a novel RSS dataset collected in front of 90 artworks across 13 museum rooms using two different platforms, i.e., Android and iOS. We provide an advanced position classification baseline taking advantage of a proximity-based method and k-NN algorithms. In our experiments, room-level accuracy ranges from 92.36 % to 99.97 % and artwork Top-3 from 75.15 % to 98.26 %, depending on the configuration, with cross-platform scenarios revealing significant challenges.

Personalization in Cultural Heritage (CH) settings is crucial for transforming visitor experiences into meaningful interactions accommodating diverse expectations and preferences.This research presents a holistic framework to enhance visitor experiences in CH physical institutions, like Galleries, Libraries, Archives, and Museums, through the combination of Indoor Positioning Systems (IPS), Recommender Systems, and Large Language Models (LLMs).Our Bluetooth beacon-based IPS implementation has been successfully deployed in a major gallery in Rome.The system covers 17 rooms and over 100 artworks and provides the user's position with high accuracy.We conceptualized a recommendation algorithm to optimize visitor engagement by progressively increasing mean dwell time while considering spatial and temporal constraints.Moreover, our experiments with LLM-generated audioguides demonstrate that visitors prefer content tailored to established visitor categories, validating our approach to personalization.These findings provide empirical support for personalized digital interpretation in GLAM contexts, though challenges remain regarding IPS precision over time and LLM hallucination mitigation.Future work will focus on collecting visitor interaction data, implementing the recommender system, and potentially releasing datasets to address the scarcity of CH-specific positioning and recommendation data.

In the last two decades, recommender systems have become more popular since they can provide personalized recommendations in different fields. However, the current research landscape in this area suggests that there is still considerable potential for applying novel recommendation techniques in indoor environments. In addition, the growing attention to privacy raises even more challenges. Anonymous session-based recommender systems represent attractive solutions in this scenario, given their natural predisposition to model the indoor domain by treating each visit to a particular location as an anonymous session. This paper presents some noteworthy challenges regarding several aspects related to the application of these models in indoor environments. We expose our research questions on issues related to the representation of user behavior, cold-start problem, and fairness. Although these problems affect any RS, they become even more challenging in the chosen environment. Finally, we outline a possible use case in a real application scenario to make more transparent and concrete the line of research we intend to pursue in the near future.