Abstract Discovering relevant datasets in large, heterogeneous data ecosystems, such as Data Lakes or Data spaces, is a complex task, often hindered by a lack of transparency and user-centric explanations in the discovery process. Explainability is critical for enabling users to understand why specific datasets are recommended, what information they contain, and how they align with user-defined criteria and preferences. To address these challenges, this work proposes a novel Graph Retrieval-Augmented Generation (Graph RAG) framework to enhance explainability in a platform for discovery of summary data sources. The proposed approach leverages a Knowledge Graph (KG) to interpret user requests, extracting relevant contextual information. These enriched requests are then transformed by a Large Language Model (LLM) into actionable dataset queries for a dataset discovery platform. Candidate solutions are evaluated and enriched with statistical insights on value distributions and contextual knowledge from the KG. Finally, the LLM ranks these solutions based on user preferences, producing a final report. This dual strategy of query enrichment and contextual explanation fosters transparency and enhances user understanding of the discovery process. We demonstrate the effectiveness of the approach through an experimental validation, highlighting its potential to improve both the accuracy and interpretability of dataset discovery.

Industry 5.0 envisions manufacturing systems that are human-centric, sustainable, and resilient. In this context, the Internet of Everything (IoE) enables integration of devices, people, and processes into a unified digital ecosystem. This paper presents a modular, semantically enriched framework that supports this transition by managing heterogeneous data sources—such as IoT sensors, wearable devices, and smart objects—through a layered architecture. The platform enables real-time data stream processing, semantic interoperability, and secure, context-aware access. Anomaly detection is enabled through a privacy-preserving mechanism based on behavioral fingerprinting and federated learning. The platform supports immersive human-machine interaction via gesture recognition, empowering workers to control and interact with industrial systems. Use cases demonstrate the system’s ability to support gesture-based control and intelligent monitoring, highlighting its potential to enhance adaptability, security, and worker empowerment in Industry 5.0 environments.

In the landscape of Industry 5.0, Internet of Everything (IoE) networks are emerging as crucial components for connecting diverse industrial sensors and devices, expanding beyond traditional IoT boundaries to integrate people, processes, and data. However, this increased connectivity raises significant security concerns, as the growing complexity of IoE environments introduces new attack vectors and privacy risks. Additionally, the integration of heterogeneous devices and data sources presents both technical and semantic interoperability challenges, requiring robust mechanisms for meaningful data interpretation and secure exchange. This paper, developed within the HOMEY project, presents an architecture for gathering and monitoring semantic data streams in IoE environments, addressing both interoperability and security challenges. Our approach leverages Knowledge Graphs to represent sensor metadata, locations, access rights, and operational contexts, enabling dynamic stream monitoring and data querying. An approach based on Federated Learning allows distributed behavioral fingerprinting of IoE devices, which is exploited on top of the platform to perform anomaly detection from real-time data streams. The approach enhances reliable, privacy-preserving anomaly detection, contributing to the security and resilience of next-generation industrial IoE ecosystems.

The use of wearable devices with inertial sensors for gesture recognition is becoming increasingly common in extended reality and remote control applications. Fast and accurate gesture recognition is critical for real-time services such as industrial control and thus drives the shift of computation toward edge computing. This work aims to optimize deep learning models for direct integration into real-world edge devices. In this context, several models based on convolutional neural networks with different configurations are developed and tested through standard performance benchmarks (accuracy and Macro F1 Score). Moreover, in order to decrease models’ size with minimal accuracy loss, they were quantized using full-integer quantization. Non-quantized and quantized models have been tested on different boards through the STM32 Edge AI Developer Cloud, as an additional benchmark to verify that the proposed models meet speed and accuracy requirements also on low-cost, low-power edge nodes. The results show that the best lightweight configuration with post-training quantization achieves an inference time of 17.31 ms on the STM32L4R9I-DISCO board, while maintaining an accuracy of 95.95 % ± 0.31 %, competitive against the state of the art and making it suitable for integration into industrial control frameworks.

Human-centric design, intelligence, and seamless interconnectivity are key pillars of the Industry 5.0. A critical challenge in these scenarios is the efficient retrieval of relevant, context-aware information for workers within Internet of Everything (IoE) networks. Traditional information retrieval techniques struggle with the heterogeneous, dynamic data generated in industrial settings. To address this, we define a context-aware data model for IoE scenarios, on top of which we propose RAG-IoE, a novel Retrieval-Augmented Generation (RAG) solution to enable adaptive, scalable, and context-based information retrieval from both structured and unstructured data sources. Our approach organizes IoE data within a semantic framework, integrating hybrid retrieval methods. It combines structured search on a Knowledge Graph with unstructured data retrieval using embeddings stored in a vector database, followed by LLM-driven reasoning to refine results. This architecture enhances decision-making, reduces cognitive overload, and ensures precise guidance for industrial operators. We validate the efficiency and effectiveness of RAG-IoE using a novel dataset through both a user study and quantitative analysis, demonstrating its potential to optimize human-machine collaboration in Industry 5.0 environments.

Recently, the Industry 5.0 is gaining attention as a novel paradigm, defining the next concrete steps toward more and more intelligent, green-aware, and user-centric digital systems. In an era in which smart devices typically adopted in the industry domain are more and more sophisticated and autonomous, the Internet of Things and its evolution, known as the Internet of Everything (IoE, for short), involving also people, robots, processes, and data in the network, represent the main driver to allow industries to put the experiences and needs of human beings at the center of their ecosystems. However, due to the extreme heterogeneity of the involved entities, their intrinsic need and capability to cooperate, and the aim to adapt to a dynamic user-centric context, special attention is required for the integration and processing of the data produced by such an IoE. This is the objective of the present paper, in which we propose a novel semantic model that formalizes the fundamental actors, elements and information of an IoE, along with their relationships. In our design, we focus on state-of-the-art design principles, in particular reuse, and abstraction, to build “SemIoE,” a lightweight ontology inheriting and extending concepts from well-known and consolidated reference ontologies. The defined semantic layer represents a core data model that can be extended to embrace any modern industrial scenario. It represents the base of an IoE knowledge graph, on the top of which, as an additional contribution, we analyze and define some essential services for an IoE-based industry.

Model repair techniques aim at automatically updating a process model to incorporate behaviors that are observed in reality but are not compliant with the original model. Most state-of-the-art techniques focus on the fitness of the repaired models, with the goal of including single anomalous behaviors observed in a log in the form of the events. This often hampers the precision of the obtained models, which end up allowing much more behaviors than intended. In the quest of techniques avoiding this over-generalization pitfall, some notion of higher-level anomalous structure is taken into account. The type of structure considered is however typically limited to sequences of low-level events. In this work, we introduce a novel repair approach targeting more general high-level anomalous structures. To do this, we exploit instance graph representations of anomalous behaviors, that can be derived from the event log and the original process model. Our experiments show that considering high-level anomalies allows to generate repaired models that incorporate the behaviors of interest while maintaining precision and simplicity closer to the original model.

Social networks are perceived by users as a natural environment for publicly sharing their thoughts and emotions about different subjects. In these platforms, despite the availability of various forms of communication, text is the most widespread way of communication. Therefore, the development of automatic techniques for emotion recognition in social media texts, such as tweets or Facebook posts, gives the opportunity of extracting information that could be valuable for many application fields, ranging from the analysis of customer satisfaction to the optimization of political campaigns. Nowadays, several emotion classifiers and datasets have been built for English texts while few resources are available for other languages. For this reason, in this work we present a deep learning algorithm for emotion recognition in Italian social media texts. The algorithm was named EmotionAlBERTo as it is based on AlBERTo, a BERT-based language understanding model for the Italian language. We trained and evaluated EmotionALBERTo on two different Twitter datasets: the MultiEmotions-It dataset and TwIT, a novel Italian dataset for emotion recognition that we built by collecting and manually labelling a corpus of about 3100 Italian tweets. Experiments show that the models achieve remarkable performance on both 4- class and 6-class emotion classification, by respectively obtaining F1= 0.91 and F1= 0.83 on MultiEmotions-It, F1= 0.92 and F1= 0.86 on TwIT.

Data Lake (DL) architectures have recently emerged as an effective solution to the problem of data analytics with big, highly heterogeneous, and quickly changing data sources. However, novel challenges arise too, including how to make sense of disparate raw data and how to identify the sources that satisfy a data need. In the paper, we introduce a semantic model for a Data Lake aimed to support data discovery and integration in data analytics scenarios. By formally modeling indicators of interest, their computation formulas, and dimensions of analysis in a knowledge graph, and by seamlessly mapping them to relevant source metadata, the framework is suited for identifying the sources and the required transformation steps according to the analytical request.

Summary A growing number of public institutions all over the world have recently started to make government statistical data available in open formats, thus enhancing transparency and accountability, stimulating innovation, and promoting civic awareness and engagement. Integration issues related to fragmentation and heterogeneity of these datasets can be partially addressed by referring to the Linked Data approach, which also enables easier access and consumption by users. However, the lack of an explicit representation of how statistical indicators are calculated still hinders their interpretation, and hence the development of applications and services especially useful for citizens, who do not have full knowledge and control over the underlying data and analysis models. In the present work, we discuss an approach to ease the interaction of communities of citizens with statistical Linked Open Data. We define a model and a set of services allowing people to recognize the mathematical structure of statistical indicators, improving in this way user awareness of the meaning of indicators and their mutual relations. Through such services, it is possible to enable interactive browsing of indicator formulas and novel typologies of data exploration, including dynamic computation of indicators not explicitly stored and comparison of different Linked Data resources.

People's daily life is increasingly intertwined with smart devices, which are more and more used in dynamic contexts. Therefore, searching and exploiting the wealth of information produced by the Internet of Things (IoT) require novel models, including a representation of the actual context of use. The definition of context is inherently difficult, due to the variety of application scenarios and user needs. In this article, we propose a general model for devices' contexts representing context components at different resolutions (or levels of granularity). This enables the definition of a multiresolution context-based algorithm for querying the IoT, according to given preferences and contexts that can be tightened or relaxed depending on the given application goal. Experimental results show how the proposed approach outperforms traditional solutions by increasing the retrieval of relevant results while keeping precision under control.

The massive adoption of social networks has made available an unprecedented amount of user-generated content, which may be analyzed in order to determine people's opinions and emotions on a large variety of topics. Research has made many efforts in defining accurate algorithms for the analysis of emotions conveyed by texts, however their performance often relies on the existence of large annotated datasets, whose current scarcity represents a major issue. The manual creation of such datasets represents a costly and time-consuming activity and hence there is an increasing demand for techniques for the automatic annotation of corpora. In this work we present a methodology for the automatic annotation of video subtitles on the basis of the analysis of facial expressions of people in videos, with the goal of creating annotated corpora that may be used to train emotion recognition algorithms. Facial expressions are analyzed through machine learning algorithms, on the basis of a set of manually -engineered facial features that are extracted from video frames. The soundness of the proposed methodology has been evaluated through an extensive experimentation aimed at determining the performance on real datasets of each methodological step.

In the last years, the wide availability on the market of low-cost smart devices paved the way for the development of smart environments, which offer an unprecedented opportunity to recognize the patterns of activities from a large amount of collected data, with the ultimate aim of monitoring the user behavior. In this article, we propose a methodology which relies on process discovery techniques to analyze sensor data in terms of activation sequences and to discover process models representing user's behavioral patterns. The extraction of such models is valuable not only in the perspective of gaining a better insight on how a certain task is performed but also in supporting novel smart services. In order to evaluate the effectiveness of the approach, in this article, we also consider a real-world case study set in an ambient-assisted living environment.

Summary Performance measurement is the subject of interdisciplinary research on information systems, organizational modeling and decision support systems. The data cube model is usually adopted to represent performance indicators (PI) and enable flexible analysis, visualization and reporting. However, the major obstacles against effective design and management of PI monitoring systems are related to the facts that PIs are complex objects with an aggregate/compound nature. This often leads to unawareness of indicator semantics as well as of dependencies among indicators. In this work, we propose to enrich the data cube model with the formal description of the structure of an indicator given in terms of its algebraic formula and aggregation function. Such a model enables the definition of a novel operator, namely indicator drill‐down, which relies on formula manipulation functionalities and reasoning. Like the usual drill‐down, this operator increases the detail of a measure of the data cube by expanding an indicator into its components. Thus, the two notions of drill‐down are integrated, allowing a novel way of data exploration. As a proof‐of‐concept, an implementation of the approach is presented. The evaluation of the implementation on real and synthetic scenarios enlightens the effectiveness and the efficiency of the approach. Copyright © 2015 John Wiley & Sons, Ltd.