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

Indoor Positioning, Localization, and Navigation (PLN) systems present unique opportunities for enhancing visitor experiences in cultural heritage venues such as museums, galleries, and exhibitions. However, cultural heritage environments impose distinct constraints such as architectural limitations, preservation requirements, and restricted sensor deployment that differentiate them from typical indoor settings. Despite growing interest in indoor PLN systems, no systematic review has examined how technologies and methodologies have been applied in these constrained cultural heritage contexts. This paper addresses this gap by presenting a systematic literature review using the PRISMA methodology of indoor PLN systems designed for, or tested in, cultural heritage venues. While PLN are commonly grouped under a single acronym, they represent distinct tasks with different goals, a distinction our review carefully accounts for in its analysis of 44 peer-reviewed works published between 2020 and 2025, examining technologies, methodologies, evaluation metrics, and their tradeoffs. Our analysis reveals significant heterogeneity across solutions, identifying eight distinct technologies and eight methodological approaches. Bluetooth emerges as the preferred technology, and important tradeoffs appear between deployment cost, accuracy, infrastructure requirements, and task-specific constraints when it comes to mixing different technologies and methodologies. Moreover, we identify severe reproducibility issues. Based on our systematic analysis, we propose practical guidelines that explicitly balance application requirements, environmental constraints, budget limitations, and accuracy expectations. We validate these guidelines through a real-world deployment using Bluetooth technology in a museum. Moreover, we highlight directions for future research, including the need for community benchmarks, overlooked technologies, standardized evaluation protocols, and improved open science practices.

Bloom’s Taxonomy plays a central role in assessment design by helping instructors align evaluation tasks with learning objectives. However, applying Bloom’s framework in practice, especially in programming education, requires substantial effort and often leads to divergent interpretations among educators. This study explores the extent to which Large Language Models (LLMs) can support the automated classification of programming assessment items across Bloom’s cognitive levels. We evaluate Bloom-based classification on a dataset of items from an introductory undergraduate Computer Science course, covering four cognitive levels (Remember, Understand, Apply, and Analyze), comparing proprietary LLMs with open-source alternatives. Our methodology considers two strategies: zero-shot prompting and a council-based ensemble approach. Results show that proprietary models achieve accuracies of up to 78 https://doi.org/10.5281/zenodo.19332485 .

Rapid progress in natural language processing (NLP) has ushered in a new era of artificial intelligence (AI) models. Large language models (LLMs), powered by deep learning algorithms, excel in understanding and generating coherent human language responses, revolutionizing various fields. In education, these models promise to enrich the learning experience by providing personalized support to students and facilitating communication between students and teachers. In particular, Generative Pre-trained Transformer (GPT) technology provides instant and contextually relevant responses, and it has become an increasingly interesting topic in education. This article explores the potential of ChatGPT to enhance learning activities in secondary schools, with a specific focus on concept maps, a valuable tool for fostering meaningful learning. Concept maps are widely recognized for their educational benefits but existing tools for automatic concept map generation lack flexibility and semantic understanding. In contrast, ChatGPT's adaptability and responsiveness make it a compelling candidate for generating concept maps tailored to individual needs. This paper sheds light on ChatGPT's potential in education and offers insights into its effectiveness in supporting learning activities through concept map generation. We present a study involving 83 secondary school students to test their perceptions of ChatGPT-generated concept maps. We compared concept maps created by both ChatGPT and teachers across six different topics. Using PlantUML, we standardized the concept maps, and then used various questionnaires to assess their quality, effectiveness, and impact on student performance. The statistical analysis revealed that the concept maps generated by ChatGPT were comparable in quality to those produced by teachers. This highlights the usefulness of ChatGPT in supporting structured learning activities while also significantly reducing the time and effort required to create concept maps.

This paper investigates the application of a Multimodal Large Language Model to enhance visitor experiences in cultural heritage settings through Visual Question Answering (VQA) and Contextual Question Answering (CQA). We evaluate the zero-shot capabilities of LLaVA-7b (Large Language and Vision Assistant) on QA using the AQUA dataset. We assess how effectively it can answer questions about artwork, visual content, and contextual information through three experimental approaches. Our findings reveal that LLaVA demonstrates promising performance on visual questions, outperforming previous baselines but facing challenges with questions requiring contextual understanding. The selective knowledge integration approach showed the best overall performance, suggesting an efficient knowledge retrieval systems could enhance performance. Moreover, we show how to exploit such models to provide correct personalized answers using a well-established visitor model.

This paper proposes using social robots to enhance children's experiences in museums. Specifically, we aim to equip these social robots with multimodal large language models (MLLMs) to generate questions that engage children interactively. To achieve this, we evaluate the capabilities of LLaVA models in generating diverse and relevant questions about artworks, comparing their performance on visual questions with contextual questions. We utilize a subset of the AQUA dataset to assess both quantitative metrics and qualitative aspects of the generated questions. Additionally, we examine the models' ability to create engaging questions tailored specifically for children. We emphasize how MLLMs can generate questions that may increase enjoyment during visits, promote active observation, and enhance children's cognitive and emotional engagement with artworks. This approach aims to contribute to more inclusive and effective learning experiences in museum settings.

Indoor Positioning Systems (IPSs) hold significant potential for enhancing visitor experiences in cultural heritage institutions. By enabling personalized navigation, efficient artifact organization, and better interaction with exhibits, IPSs can transform the modalities of 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. In other contexts, 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. Additionally, we provide an advanced position classification baseline taking advantage of a proximity-based method and k-NN algorithms. In our analysis, we discuss the results and offer suggestions for potential research directions.

Code assessment in computer science education is a time-consuming process requiring evaluation of both functionality and coding style. To overcome this problems, we present a novel approach that utilizes Large Language Models (LLMs) for rubric-based code evaluation, addressing the limitations of current automated tools that rely primarily on test cases and code similarity metrics, which sometimes fail to capture more nuanced aspects of student submissions. Our method enables LLMs to analyze student code against instructor-defined rubrics, generating inline, contextualized feedback without altering the original code. This may allow instructors to efficiently review, refine, and finalize grades while maintaining pedagogical oversight. We evaluated the grading performance of five state-of-the-art LLMs (Claude-3.5 Sonnet, GPT-4o, Grok-2, LLaMA-3.3 70B, and DeepSeek-V3) against human graders using a dataset of over 500 coding exercises from an introductory university programming course. Our findings show that Claude-3.5 Sonnet exhibits the highest alignment with human grading, consistently outperforming other models across different exercise types. The dataset and code for reproducibility are available at: https://doi.org/10.5281/zenodo.14879148.

Integrating Artificial Intelligence (AI) in educational settings has brought new learning approaches, transforming the practices of both students and educators. Among the various technologies driving this transformation, Large Language Models (LLMs) have emerged as powerful tools for creating educational materials and question answering, but there are still space for new applications. Educators commonly use Multiple-Choice Questions (MCQs) to assess student knowledge, but manually generating these questions is resource-intensive and requires significant time and cognitive effort. In our opinion, LLMs offer a promising solution to these challenges. This paper presents a novel comparative analysis of three widely known LLMs - Llama 2, Mistral, and GPT-3.5 - to explore their potential for creating informative and challenging MCQs. In our approach, we do not rely on the knowledge of the LLM, but we inject the knowledge into the prompt to contrast the hallucinations, giving the educators control over the test’s source text, too. Our experiment involving 21 educators shows that GPT-3.5 generates the most effective MCQs across several known metrics. Additionally, it shows that there is still some reluctance to adopt AI in the educational field. This study sheds light on the potential of LLMs to generate MCQs and improve the educational experience, providing valuable insights for the future.

Eye-tracking technology has gained prominence in cultural heritage studies, facilitating behavioral analysis and visitor engagement assessments. This paper explores the challenges and future directions of artwork segmentation in eye-tracking experiments, aiming to automate the identification of areas of interest. Although existing segmentation approaches, such as semantic segmentation models, show promise, they face limitations in accurately segmenting diverse artwork styles. We propose hybrid segmentation as a viable strategy, combining multiple techniques for improved accuracy. Through qualitative analysis, we evaluate segmentation models on public domain artworks, highlighting the strengths and weaknesses of each approach.

Recommender systems (RSs) are increasingly present in our everyday lives for business and pleasure. The Cultural Heritage domain is no exception. In the research literature, several RSs have been proposed to enhance the fruition of artistic and cultural resources. In this paper, we present some of our research activities aimed at realizing a RS for suggesting personalized itineraries to exhibit and museum visitors. More specifically, we describe the collection and use of eye-tracking data to understand if there are any correlations between the visitors’ gaze patterns and their degree of appreciation of the viewed artworks. If such correlations exist, they could be used as implicit feedback in the recommendation engine. The preliminary results are interesting and encourage us to pursue our research activities.

Emotion modeling for social robotics has the great potential to improve the life quality for the elderly and individuals with disabilities by making communication, care, and interactions more effective. It can help individuals with communication difficulties express their emotions. It can also be used to monitor the emotional well-being of elderly persons living alone and alert caregivers or family members if there are signs of distress. More broadly, emotion modeling is necessary to design robots closer and closer to human beings that can naturally interact with them by understanding their behavior and reactions. Here, we propose a deep learning technique for emotion classification using electroencephalogram (EEG) signals. We aim to recognize valence, arousal, dominance, and likability. Our technique uses the spectrogram from each of the 32 electrodes applied in the skull area. Then, we employ a Resnet101 convolutional neural network to learn a model capable of predicting several emotions. We built and tested our model on the DEAP dataset.

Although personalization is a staple in several online settings, achieving an ad-hoc experience in some environments is impossible based on personal tastes. One such environment is the museum. In our view, visitors’ facial reactions in front of artworks can play a crucial role. In this context, we want to study visitor behavior with an even finer-grained approach, identifying the most activated brain areas and how they relate to facial expressions. This paper describes how we intend to create a multimodal dataset to validate our study. We aim to fill a gap in personalizing the heritage experience with multidisciplinary research that combines neuroscience and computer science

The Web offers an unprecedented number of resources and has become the most popular source of information for students shaping their understanding of a new topic, and for instructors selecting relevant material for learning and teaching activities. Even though search engines are the most widely used tools for searching for educational content, the realities of the learning and teaching processes make the retrieval and evaluation of educational resources more complex than they are for other goods or services. The lack of recourse to educational metadata in web pages, as well as the size of the Web itself, call for specific techniques to be adopted for a more effective ranking of educational content. In this study, we propose an innovative approach based on semantic technologies. The SemanticSearch approach described in this paper leverages knowledge graph representation of teaching contexts and proposes a new ranking method for rating educational web content. In the literature we find an Educational Ranking Principle that ranks web pages for a specific teaching context. In this study, we integrate the Educational Ranking Principle with semantic data to extend the experimentation and analyse performance further. We undertake an evaluation involving university teachers, considering more than 70 queries to measure the SemanticSearch performance against the Educational Ranking Principle in addition to two state-of-the-art methodologies: Tf-Idf and BM25F. Paired t-tests of four accuracy measures provide statistical evidence for improvements made by using SemanticSearch method when compared to the three baselines.

In this paper, we present the rationale and the ideas behind META4RS, a museum itinerary recommender system. The system leverages deep learning techniques to acquire data about the visitor’s position while ensuring her anonymity. Moreover, the visitor’s appraisal of the artwork she observes is inferred implicitly based on the emotional reactions she expresses while watching a given artwork. We are not aware of any such recommender system proposed in the research literature. However, this system should ensure several advantages: (i) it is non-intrusive since it makes use of simple badges and off-the-shelf cameras while ensuring the anonymity of the visitor; (ii) it is independent of the type of museum; (iii) it offers personalized itineraries to visitors based on their implicitly inferred interests and preferences. Specifically, we illustrate the background and describe the architecture of the proposed system, discussing the steps required for its implementation. We also provide details of what has already been done and what remains to be done, outlining the open problems.

Nowadays, technology makes it possible to admire objects and artworks exhibited all over the world remotely. We have been able to appreciate this convenience even more in the last period, in which the pandemic has forced us into our homes for a long time. However, visiting art sites in person remains a truly unique experience. Even during on-site visits, technology can help make them much more satisfactory, by assisting visitors during the fruition of cultural and artistic resources. To this aim, it is necessary to monitor the active user for acquiring information about their behavior. We, therefore, need systems able to monitor and analyze visitor behavior. The literature proposes several techniques for the timing and tracking of museum visitors. In this article, we propose a novel approach to indoor tracking that can represent a promising and non-expensive solution for some of the critical issues that remain. In particular, the system we propose relies on low-cost equipment (i.e., simple badges and off-the-shelf RGB cameras) and harnesses one of the most recent deep neural networks (i.e., Faster R-CNN) for detecting specific objects in an image or a video sequence with high accuracy. An experimental evaluation performed in a real scenario, namely, the “Exhibition of Fake Art” at Roma Tre University, allowed us to test our system on site. The collected data has proven to be accurate and helpful for gathering insightful information on visitor behavior.

Although ubiquitous and fast access to the Internet allows us to admire objects and artworks exhibited worldwide from the comfort of our home, visiting a museum or an exhibition remains an essential experience today. Current technologies can help make that experience even more satisfying. For instance, they can assist the user during the visit, personalizing her experience by suggesting the artworks of her higher interest and providing her with related textual and multimedia content. To this aim, it is necessary to automatically acquire information relating to the active user. In this paper, we show how a deep neural network-based approach can allow us to obtain accurate information for understanding the behavior of the visitor alone or in a group. This information can also be used to identify users similar to the active one to suggest not only personalized itineraries but also possible visiting companions for promoting the museum as a vehicle for social and cultural inclusion.