Generative agents are rapidly advancing in sophistication, raising urgent questions about how they might coordinate when deployed in online ecosystems. This is particularly consequential in information operations (IOs), influence campaigns that aim to manipulate public opinion on social media. While traditional IOs have been orchestrated by human operators and relied on manually crafted tactics, agentic AI promises to make campaigns more automated, adaptive, and difficult to detect. This work presents the first systematic study of emergent coordination among generative agents in simulated IO campaigns. Using generative agent-based modeling, we instantiate IO and organic agents in a simulated environment and evaluate coordination across operational regimes, from simple goal alignment to team knowledge and collective decision-making. As operational regimes become more structured, IO networks become denser and more clustered, interactions more reciprocal and positive, narratives more homogeneous, amplification more synchronized, and hashtag adoption faster and more sustained. Remarkably, simply revealing to agents which other agents share their goals can produce coordination levels nearly equivalent to those achieved through explicit deliberation and collective voting. Overall, we show that generative agents, even without human guidance, can reproduce coordination strategies characteristic of real-world IOs, underscoring the societal risks posed by increasingly automated, self-organizing IOs.

Cyber-Physical Systems (CPS) drive modern critical infrastructures by tightly integrating physical processes with computation and communication networks. This convergence exposes CPS to sophisticated cyber threats propagating across physical, control, and network layers, where stealthy and multi-stage attacks manifest through weak and distributed signals, challenging centralized intrusion detection systems and monolithic AI models that struggle to scale across heterogeneous subsystems and lack the interpretability required in safety-critical environments. This paper proposes a hierarchical multi-agent AI framework for automated cybersecurity assessment with human-understandable explanations in networked CPS. The framework dynamically instantiates a task-specific agent hierarchy from a natural language description of the target system, aligning the security analysis process with the underlying CPS architecture. Specialized agents perform fine-grained analysis of heterogeneous data sources, subsystem supervisors aggregate and contextualize local findings, and a deliberative round-table consensus mechanism enables cross-subsystem correlation for detecting coordinated and stealthy attacks. The framework is evaluated using four LLMs (Qwen 3 4B, Qwen 3 8B, Llama 3.1 8B, and Llama 3.3 70B) across three datasets (PicoDomain, CERT r5.2, and SWaT), achieving perfect recall for models with 8B parameters and above (i.e., zero missed attacks), while reaching an accuracy of 90.18% on PicoDomain, 92.77% on CERT r5.2, and 96.97% on SWaT. An ablation study confirms the effectiveness of the cross-subsystem consensus mechanism, demonstrating substantial precision gains when collaborative deliberation is enabled. Overall, this work establishes hierarchical multi-agent architectures as a scalable, interpretable, and structure-aware foundation for AI-driven cybersecurity in networked CPS.

Abstract Named Entity Recognition (NER) in specialized domains like biomedicine suffers from acute data scarcity, requiring expensive expert annotations. While data augmentation offers a promising solution, it inevitably introduces noisy and mislabeled samples that can degrade model performance. This problem is amplified in few-shot scenarios where every training example matters. We introduce PALAUNER (Policy-based Active Learning to Augment Named Entity Recognition), a reinforcement learning framework that learns to select high-quality samples from augmented data pools. Using a deep Q-network, our agent evaluates samples based on content features and model predictions, deciding which examples will improve NER performance. Experiments across five BioNER benchmarks demonstrate that PALAUNER consistently enhances diverse augmentation methods, from simple perturbations to GPT-based generation. Average F1 improvements are of 0.5−7.1 points in few-shot settings. PALAUNER’s modular design enables seamless integration with emerging augmentation techniques, providing a generalizable solution for training data quality enhancement. We publicly release our code on GitHub: ( https://github.com/picuslab/palauner ).

Abstract Smart cities, as a constantly evolving field, offer numerous opportunities for advancement and innovation. In particular, more and more specific sectors within smart cities currently require increased attention and development. Notably, the integration of emerging technologies holds promise for significant progress in efficiency and management across various urban issues, with urban waste management being a key focus of this research. In this paper, we propose a new methodology for intelligent waste management combining graph-based modeling and advanced machine learning techniques. A case study of a municipality in the Melbourne area (Australia) is examined, where smart bins equipped with fill-level sensors have been deployed, and their data made publicly accessible. The core of this investigation lies in utilizing sophisticated machine learning models, including Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), for predicting bin fill levels. Additionally, we introduce a graph-based spatial representation in which bins are modeled as nodes of a weighted graph, and graph embedding techniques are employed to capture and encode spatial dependencies among them. The findings validate the effectiveness of the proposed methodology, achieving a Mean Absolute Error (MAE) of 0.16, a Root Mean Squared Error (RMSE) of 0.21, a Mean Squared Error (MSE) of 0.044, and an R 2 of 0.83, which outperforms the reference baseline. Our results highlight how the integration of spatial information through graph embedding holds promise for improving predictions in smart waste management. Furthermore, the study suggests a transferable methodological framework, applicable at the architectural level to smart city challenges where spatially distributed entities exhibit interdependent temporal dynamics, pending empirical validation across further domains and cities.

Vision-Language Models (VLMs) excel in multimodal tasks, yet their effectiveness in specialized medical applications remains underexplored. Accurate interpretation of medical images and text is crucial for clinical decision support, particularly in multiple-choice question answering (MCQA). To address the lack of benchmarks in this domain, we introduce the Multilingual Multimodal Medical Exam Dataset (MMMED), designed to assess VLMs' ability to integrate visual and textual information for medical reasoning. MMMED includes 582 MCQA pairs from Spanish medical residency exams (MIR), with multilingual support (Spanish, English, Italian) and paired medical images. We benchmark state-of-theart VLMs, analyzing their strengths and limitations across languages and modalities. The dataset is publicly available on Hugging Face (https://huggingface.co/datasets/praiselab-picuslab/MMMED), with experimental code on GitHub (https://github.com/PRAISELab-PicusLab/MMMED).

Medical question answering systems face deployment challenges including hallucinations, bias, computational demands, privacy concerns, and the need for specialized expertise across diverse domains. Here, we present SOLVE-Med, a multi-agent architecture combining domain-specialized small language models for complex medical queries. The system employs a Router Agent for dynamic specialist selection, ten specialized models (1B parameters each) fine-tuned on specific medical domains, and an Orchestrator Agent that synthesizes responses. Evaluated on Italian medical forum data across ten specialties, SOLVE-Med achieves superior performance with ROUGE-1 of 0.301 and BERTScore F1 of 0.697, outperforming standalone models up to 14B parameters while enabling local deployment. Our code is publicly available on GitHub: https://github.com/PRAISELab-PicusLab/SOLVE-Med.

Artificial Intelligence (AI) systems are transforming critical sectors such as healthcare, finance, and transportation, enhancing operational efficiency and decision-making processes. However, their deployment in high-stakes domains has exposed vulnerabilities that can result in significant societal harm. To systematically study and mitigate these risk, initiatives like the AI Incident Database (AIID) have emerged, cataloging over 3,000 real-world AI failure reports. Currently, associating a new report with the appropriate AI Incident relies on manual expert intervention, limiting scalability and delaying the identification of emerging failure patterns. To address this limitation, we propose a retrieval-based framework that automates the association of new reports with existing AI Incidents through semantic similarity modeling. We formalize the task as a ranking problem, where each report—comprising a title and a full textual description—is compared to previously documented AI Incidents based on embedding cosine similarity. Benchmarking traditional lexical methods, cross-encoder architectures, and transformer-based sentence embedding models, we find that the latter consistently achieve superior performance. Our analysis further shows that combining titles and descriptions yields substantial improvements in ranking accuracy compared to using titles alone. Moreover, retrieval performance remains stable across variations in description length, highlighting the robustness of the framework. Finally, we find that retrieval performance consistently improves as the training set expands. Our approach provides a scalable and efficient solution for supporting the maintenance of the AIID.

The rapid rise of deep generative models has enabled realistic synthetic media, posing severe risks to healthcare communication where trust and accuracy are essential. Existing deepfake detectors, mostly trained on entertainment-oriented datasets, degrade sharply in medical contexts, where domainspecific terminology and evidence-based discourse are critical. Moreover, post-hoc moderation is often too slow to mitigate the spread of health-related disinformation, underscoring the need for timely, domain-adapted solutions. We address these gaps by (i) curating a multimodal healthcare-oriented corpus of in-the wild and synthetically generated deepfakes, (ii) benchmarking state-of-the-art monomodal detectors for video, audio, and text, and (iii) introducing a late-fusion framework that integrates their predictions via ensemble methods. Our experiments show that late fusion consistently outperforms individual detectors, yielding notable gains in macro-F1 and AUC (+0.159 and +0.131, respectively), particularly under domain-shift and lowresource conditions. These findings demonstrate the importance of multimodal integration for robust detection of health-related deepfakes and contribute toward building trustworthy AI systems to safeguard medical communication. Code and data will be released for transparency and reproducibility: https://github.com/PRAISELab-PicusLab/Deepfake-Detection-in-Healthcare

The increasing volume of unstructured clinical data—such as consultation audio, notes, and reports—has intensified the documentation burden on physicians, limiting time for direct patient interaction. Existing transcription and summarization systems remain largely proprietary and opaque, offering limited semantic interpretability or clinical control. We present AMICO (Anamnesi Medica Intelligente del Colloquio Ospedaliero), a multimodal framework for AI-assisted clinical documentation that integrates audio acquisition, semantic processing, and interactive human validation. The system combines automatic speech recognition, medical named entity recognition, summarization, and drug-drug interaction checking within a unified workflow designed to support, not replace, physician expertise. Evaluation shows that a fine-tuned LLM achieves superior accuracy and faithfulness compared to baseline LLMs, confirming the effectiveness of domain-specific adaptation for Italian medical dialogue summarization. AMICO demonstrates that moderately sized, transparent, and human-supervised models can meaningfully reduce documentation workload while ensuring clinical reliability and safety.

Harmful meme detection poses a critical challenge for online moderation, as the multimodal and context-dependent nature of memes undermines the effectiveness of traditional unimodal classifiers. In this work, we propose an agent-driven architecture that combines multimodal decomposition with multiagent reasoning. The system extracts complementary information from memes through text extraction, image captioning, and contextual visual description, which are integrated into a unified textual representation. This representation is then analyzed by a set of LLM-powered agents, each instantiated with a distinct interpretative persona, whose assessments are consolidated by a decision aggregation module. Experimental results on the Facebook Hateful Memes dataset demonstrate that the multiagent approach significantly improves performance: accuracy increases from$5 4. 2 5 \%$(single-agent baseline) to$6 7. 2 8 \%$, while the true positive rate rises from 40.32% to 72.78%. These findings highlight the effectiveness of integrating multimodal decomposition with agent-based perspectives for harmful meme detection, ensuring both robustness and explainability in the decision-making process.

Misinformation in healthcare, from vaccine hesitancy to unproven treatments, poses risks to public health and trust in medical systems. While machine learning and natural language processing have advanced automated fact-checking, validating biomedical claims remains uniquely challenging due to complex terminology, the need for domain expertise, and the critical importance of grounding in scientific evidence. We introduce CER (Combining Evidence and Reasoning), a novel framework for biomedical fact-checking that integrates scientific evidence retrieval, reasoning via large language models, and supervised veracity prediction. By integrating the text-generation capabilities of large language models with advanced retrieval techniques for high-quality biomedical scientific evidence, CER effectively mitigates the risk of hallucinations, ensuring that generated outputs are grounded in verifiable, evidence-based sources. Evaluations on expert-annotated datasets (HealthFC, BioASQ-7b, SciFact) demonstrate state-of-the-art performance and promising cross-dataset generalization. Code and data are released for transparency and reproducibility: https://github.com/PRAISELab-PicusLab/CER

Biomedical misinformation - ranging from misleading social media posts to fake news articles and deepfake videos - is increasingly pervasive across digital platforms, posing significant risks to public health and clinical decision-making. We developed CER, a comprehensive fact-checking system designed specifically for biomedical content. CER integrates specialized components for claim detection, scientific evidence retrieval, and veracity assessment, leveraging both transformer models and large language models to process textual and multimedia content. Unlike existing fact-checking systems that focus solely on structured text, CER can analyze claims from diverse sources including videos and web content through automatic transcription and text extraction. The system interfaces with PubMed for evidence retrieval, employing both sparse and dense retrieval methods to gather relevant scientific literature. Our evaluations on standard benchmarks including HealthFC, BioASQ-7, and SciFact demonstrate that CER achieves state-of-the-art performance, with F1-score improvements compared to existing approaches. To ensure reproducibility and transparency, we release the GitHub repository with the source code, within which you can reach an interactive demonstration of the system published on HuggingFace and a video demonstration of the system \https://github.com/PRAISELab-PicusLab/CER-Fact-Checking.

The spread of misinformation and fake news on digital platforms poses significant societal challenges, underscoring the need for robust detection. Multi-task learning leverages relationships among disinformation-related tasks (e.g., stance detection, rumor classification) to enhance detection; however, it risks negative transfer , potentially degrading performance instead of achieving positive transfer . In this paper, we systematically investigate the mechanisms underlying positive and negative transfers across a comprehensive set of disinformation-related tasks, including Sentiment Analysis (SA), Fake News Detection (FND), Stance Detection (SD), and Topic Detection (TD). Specifically, we pioneer the use of explanations to uncover the differences between models trained under single-task and multi-task settings. Our results reveal instances of positive transfer across several task combinations, with multi-task learning yielding performance improvements of 3.26%, 6.57%, and 0.62% for SA, FND and TD tasks, respectively. Furthermore, when comparing explanations of single-task and multi-task models, we find that positive transfer refines the knowledge that can already be learnt in single-task settings by incorporating additional patterns from other tasks. Conversely, negative transfer significantly undermines models' knowledge to the extent that their explanations are equivalent to a random perturbation of the explanations generated by their single-task counterparts.

Geopolitical news provides vast amounts of information essential for understanding international relations and political events. However, organizing this information into a coherent, structured format poses challenges due to the complexity and dynamic nature of the domain. This paper introduces a scalable system leveraging Large Language Models to build continuously updated Knowledge Graphs from Italian geopolitical news. The system features a modular architecture, including a Collector Node for scalable article extraction, a Redis-based reliable queue to manage large-scale data ingestion, and a Named Entity Recognition/Relation Extraction Engine to standardize entity-relation triples. The framework addresses key challenges, such as continuous updating and hallucination mitigation, ensuring the reliability of the graph. Our evaluations demonstrate significant improvements in scalability, uniformity of extracted triples, and graph accuracy, making this architecture particularly suitable for real-time geopolitical analysis.

Legal Artificial Intelligence has emerged as an essential field, focusing on AI technologies that facilitate various legal tasks and alleviate the workload of legal professionals. Despite advancements in Legal Artificial Intelligence, there remains a critical gap in systems that can provide both comprehensive and contextually accurate retrieval tailored to the intricate structure of EU legislation. We propose EuropeanLawAdvisor, an efficient and user-friendly legal information retrieval system designed to deliver tailored responses to legal queries. This system utilizes open-source Large Language Models within a Retrieval-Augmented Generation framework, facilitating precise and relevant information retrieval. The system employs a robust retrieval approach that integrates multi-match, k-nearest neighbors, hybrid methods, and TF-IDF search strategies across both complete documents and segmented text indexes, ensuring comprehensive retrieval for diverse query types. The implementation of the framework has demonstrated significant improvements in the accuracy and relevance of responses to EU legal queries, enhancing both the retrieval of relevant legal documents and the generation of precise responses. We show that EuropeanLawAdvisor, leveraging open-source models like Phi3-mini-3B and LLaMa-3-8B, achieves competitive Faithfulness and Relevance compared to GPT-4-Turbo. The performance gap narrows significantly in zero-shot scenarios, and our approach outperforms GPT-4-Turbo in the percentage of answered questions. We publicly release our code on GitHub: https://github.com/raffaele-russo/EuropeanLawAdvisor.

Nowadays, our security and privacy are strongly threatened by malware programs which aim to steal our confidential data and make our systems out of service, among other things. While traditional signature-based malware detection methods or statistical analysis have proven to be ineffective and time-consuming, recently data-driven Artificial Intelligence (AI) techniques, i.e. Machine Learning (ML) and Deep Learning (DL) approaches, have been successfully applied leveraging the behaviour of malware in terms of API calls, and achieving promising performances. However, their black-box behavior leads to a lack of explainability thus preventing their application in real world scenarios. In light of this, eXplainable Artificial Intelligence (XAI) methodologies and tools can be effectively embedded within an AI-based malware detection process in order to make more understandable the produced results. In this paper, we propose a XAI framework for behavioral malware detection problems and evaluate the usefulness of four XAI methods (SHAP, LIME, LRP and Attention mechanism) on three datasets with different size, sequence length and number of classes, by which we could evaluate the strengths and weaknesses – from effectiveness and efficiency point of views – of recurrent deep architectures (i.e. Long-Short Term Memory (LSTM) and Gated Recurrent Unit (GRU) models), and their applicability in the modern Cyber Security (CS) scenarios.

Despite the vast potential for insights and value present in Electronic Health Records (EHRs), it is challenging to fully leverage all the available information, particularly that contained in the free-text data written by clinicians describing the health status of patients. The utilization of Named Entity Recognition and Linking tools allows not only for the structuring of information contained within free-text data, but also for the integration with medical ontologies, which may prove highly beneficial for the analysis of patient medical histories with the aim of forecasting future medical outcomes, such as the diagnosis of a new disorder. In this paper, we propose MedTKG, a Temporal Knowledge Graph (TKG) framework that incorporates both the dynamic information of patient clinical histories and the static information of medical ontologies. The TKG is used to model a medical history as a series of snapshots at different points in time, effectively capturing the dynamic nature of the patient's health status, while a static graph is used to model the hierarchies of concepts extracted from domain ontologies. The proposed method aims to predict future disorders by identifying missing objects in the quadruple 〈s, r, ?, t 〉, where s and r denote the patient and the disorder relation type, respectively, and t is the timestamp of the query. The method is evaluated on clinical notes extracted from MIMIC-III and demonstrates the effectiveness of the TKG framework in predicting future disorders and of medical ontologies in improving its performance.

In this paper, we address issues associated with the effective management of handover events in satellite-enabled 5G network infrastructures. Namely, we devise a strategy for dynamically allocating 5G gNB available resources in the presence of a constellation of LEO satellites, based on several parameters collected and dispatched by an ad hoc orchestration platform. We propose to leverage a Combinatorial Multi-Armed Bandit approach to design a resource allocation game that helps make decisions over time under uncertainty conditions related to the incidence of several factors that can determine the quality of experience perceived on the user equipment side. The introduced approach does represent a pioneering one, since it allows us to model a joint optimization task as a competitive game in which agents typically share resources with other agents instead of occupying them exclusively. The designed task allows to dynamically enable and disable channels, taking care of the relationships with the lower layer, transparently managing the required handover operations, and considering possible interference with other channels. We discuss an implementation of the proposed solution in a simulated environment. We also analyze the performance it attains, by measuring both its efficiency and efficacy in a trial setup reproducing a scenario with near real-time temporal requirements. Results show that the proposed approach has a linear trend in terms of running time with respect to the number of user equipments and gNbs involved, while achieving a sub-optimal solution of the handover task in around 20–30 rounds.

Deep learning, owing to its transformative influence across a myriad of sectors, has recently made its foray into the legal domain, instigated by the surge in digitization. Among the multitude of applications in this space, legal document classification emerges as a pivotal yet complex undertaking. Legal texts, characterized by unique domain-centric semantics and intricate linguistic patterns, necessitate precision-driven classification systems for numerous practical implications. This paper illuminates the challenges and opportunities in automating the classification of European Union (EU) legal documents, emphasizing the interrelationships among statutes and the hierarchical nature of legal references. In this context, we introduce a novel graph data modeling technique that adeptly marries content-centric indicators with the relational dynamics inherent among diverse legal documents. Central to our approach is a framework that melds text embeddings with graph neural networks for the classification of legal documents aligned with their subject-based directories. Empirical evaluations on the EU law dataset underline the efficacy of our model across varying granularities, from general thematic categories to intricate subtopics. This endeavor not only augments the comprehensibility and accessibility of EU jurisprudence but also holds significant implications across regulatory compliance, legal research, and policy formulation, underscoring the potential of deep learning in reshaping legal paradigms.

In the railway domain, rolling stock maintenance affects service operation time and efficiency. Minimizing train unavailability is essential for reducing capital loss and operational costs. To this aim, prediction of failures of rolling stock equipment is crucial to proactively trigger proper maintenance activities. Indeed, predictive maintenance is a golden example of the digital transformation within Industry 4.0, which affects several engineering processes in the railway domain. Nowadays, it may leverage artificial intelligence and machine learning algorithms to forecast failures and schedule the optimal time for maintenance actions. Generally, rail systems deteriorate gradually over time or fail directly, leading to data that vary extremely slowly. Indeed, ML approaches for predictive maintenance should consider this type of data to accurately predict and forecast failures. This paper proposes a methodology based on Long Short-Term Memory deep learning algorithms for predictive maintenance of railway rolling stock equipment. The methodology allows us to properly learn long-term dependencies for gradually changing data, and both predicting and forecasting failures of rail equipment. In the framework of an academic-industrial partnership, the methodology is experimented on a train traction converter cooling system, demonstrating its applicability and benefits. The results show that it outperforms state-of-the-art methods, reaching a failure prediction and forecasting accuracy over 99%, with a false alarm rate of ∼0.4% and a mean absolute error in the order of 10−4, respectively.

Although in the last decade several fact-checking organizations have emerged to verify misinformation, fake news has continued to proliferate, especially through social media platforms. Even though adopting improved detection strategies is of utmost importance, the fact-checking process could be optimised by verifying whether a claim has been previously fact-checked. Despite some ad-hoc information retrieval approaches having been recently proposed, the utility of modern (neural) retrieval systems have not been investigated yet. In this paper, we consider the standard two-phases retriever-reranker architecture and benchmark different state-of-the-art techniques from the information retrieval and Q&A literature. We design several experiments on a real-world Twitter dataset to analyse the efficiency and the effectiveness of the benchmark approaches. Our results show that combining standard and neural approaches is the most promising research direction to improve retrievers performance and that complex (neural) rerankers might still be efficient in practice since there is no need to process a high number of documents to improve ranking performance.

Recent years have seen an exponential growth (+98% in 2022 w.r.t. the previous year) of the number of research articles in the few-shot learning field, which aims at training machine learning models with extremely limited available data. The research interest toward few-shot learning systems for Named Entity Recognition (NER) is thus at the same time increasing. NER consists in identifying mentions of pre-defined entities from unstructured text, and serves as a fundamental step in many downstream tasks, such as the construction of Knowledge Graphs, or Question Answering. The need for a NER system able to be trained with few-annotated examples comes in all its urgency in domains where the annotation process requires time, knowledge and expertise (e.g., healthcare, finance, legal), and in low-resource languages. In this survey, starting from a clear definition and description of the few-shot NER (FS-NER) problem, we take stock of the current state-of-the-art and propose a taxonomy which divides algorithms in two macro-categories according to the underlying mechanisms: model-centric and data-centric. For each category, we line-up works as a story to show how the field is moving toward new research directions. Eventually, techniques, limitations, and key aspects are deeply analyzed to facilitate future studies.

In Biomedical Named Entity Recognition (BioNER), the use of current cutting-edge deep learning-based methods, such as deep bidirectional transformers (e.g. BERT, GPT-3), can be substantially hampered by the absence of publicly accessible annotated datasets. When the BioNER system is required to annotate multiple entity types, various challenges arise because the majority of current publicly available datasets contain annotations for just one entity type: for example, mentions of disease entities may not be annotated in a dataset specialized in the recognition ofdrugs, resulting in a poor ground truth when using the two datasets to train a single multi-task model. In this work, we proposeTaughtNet, a knowledge distillationbased framework allowing us to fine-tune a single multi-task student model by leveraging both the ground truth and the knowledge of single-taskteachers. Our experiments on the recognition of mentions of diseases, chemical compounds and genes show the appropriateness and relevance of our approach w.r.t. strong state-of-the-art baselines in terms of precision, recall and F1 scores. Moreover, TaughtNet allows us to train smaller and lighter student models, which may be easier to be used in real-world scenarios, where they have to be deployed on limitedmemory hardware devices and guarantee fast inferences, and shows a high potential to provide explainability. We publicly release both our code on github1 and our multi-task model on the huggingface repository.

In recent years, we have witnessed an ever wider spread of multimedia streaming platforms (e.g., Netflix, Spotify, and Amazon). Hence, it has become more and more essential to provide such systems with advanced recommendation facilities, in order to support users in browsing these massive collections of multimedia data according to their preferences and needs. In this context, the modeling of entities and their complex relationships (e.g., users listening to topic-based songs or authors creating different releases of their lyrics) represents the key challenge to improve the recommendation and maximize the users' satisfaction. To this end, this is the first study to leverage the high representative power of hypergraph data structures in combination with modern graph machine learning techniques in the context of music recommendation. Specifically, we propose hypergraph embeddings for music recommendation (HEMR), a novel framework for song recommendation based on hypergraph embedding. The hypergraph data model allows us to represent seamlessly all the possible and complex interactions between users and songs with the related characteristics; meanwhile, embedding techniques provide a powerful way to infer the user-song similarities by vector mapping. We have experimented the effectiveness and efficiency of our approach with respect to the state-of-the-art most recent music recommender systems, exploiting the Million Song dataset. The results show that HEMR significantly outperforms other state-of-the-art techniques, especially in scenarios where the cold-start problem arises, thus making our system a suitable solution to embed within a music streaming platform.

Federated learning has been increasingly studied to cope with the scalability and privacy issues characterizing current and upcoming large-scale infrastructures, such as the Internet of Things, 5G networks, vehicular applications, and so on. This approach partitions the data storage and AI model training in a series of local learners, whose results are aggregated by a central server and then redistributed back to the learners to have a trained global model. However, despite avoiding outsourcing sensitive data to cloud-hosted services and fragmenting the workload for data processing, such a decentralized learning approach lays the overall solution open to various kinds of attacks, able to fully compromise the accuracy of the obtained global model. This study aims to quantitatively assess the impact of two widely-recognized attacks against federated learning and propose a tentative protection means by using blockchain.

Nowadays, really huge volumes of fake news are continuously posted by malicious users with fraudulent goals thus leading to very negative social effects on individuals and society and causing continuous threats to democracy, justice, and public trust. This is particularly relevant in social media platforms (e.g., Facebook, Twitter, Snapchat), due to their intrinsic uncontrolled publishing mechanisms. This problem has significantly driven the effort of both academia and industries for developing more accurate fake news detection strategies: early detection of fake news is crucial. Unfortunately, the availability of information about news propagation is limited. In this paper, we provided a benchmark framework in order to analyze and discuss the most widely used and promising machine/deep learning techniques for fake news detection, also exploiting different features combinations w.r.t. the ones proposed in the literature. Experiments conducted on well-known and widely used real-world datasets show advantages and drawbacks in terms of accuracy and efficiency for the considered approaches, even in the case of limited content information.

Hard disk drive failures are one of the most common causes of service downtime in data centers. Predictive maintenance techniques have been adopted to extend the Remaining Useful Life (RUL) of these drives, and minimize service shortage and data loss. Several approaches based on machine and deep learning techniques have been proposed to address these issues, mostly exploiting models based on Self-Monitoring analysis and Reporting Technology (SMART) attributes. While these models have proven to be reliable, their performance is affected by the lack of information about the proximity of disk failure in time. Moreover, many of these techniques are sensitive to the highly unbalanced nature of existing data-sets, in terms of good to failed hard disk ratio. In this article, we propose a LSTM (Long Short Term Memory)-based model combining SMART attributes and temporal analysis for estimating a hard drive health status according to its time to failure. Our approach outperforms state-of-the-art methods when evaluated on two data-sets, one containing hourly samples from 23395 disks and the other reporting daily samples from 29878‬ disks. Experimental results showed that our approach is well suited to data-sets with different sampling periods, being able to predict hard drive health status up to 45 days before failure.

Social Networks represent a cornerstone of our daily life, where the so-called social reviewing systems (SRSs) play a key role in our daily lives and are used to access data typically in the form of reviews. Due to their importance, social networks must be trustworthy and secure, so that their shared information can be used by the people without any concerns, and must be protected against possible attacks and misuses. One of the most critical attacks against the reputation system is represented by mendacious reviews. As this kind of attacks can be conducted by legitimate users of the network, a particularly powerful solution is to exploit trust management, by assigning a trust degree to users, so that people can weigh the gathered data based on such trust degrees. Trust management within the context of SRSs is particularly challenging, as determining incorrect behaviors is subjective and hard to be fully automatized. Several attempts in the current literature have been proposed; however, such an issue is still far from been completely resolved. In this study, we propose a solution against mendacious reviews that combines fuzzy logic and the theory of evidence by modeling trust management as a multicriteria multiexpert decision making and exploiting the novel concept of time-dependent and content-dependent crown consensus. We empirically proved that our approach outperforms the main related works approaches, also in dealing with sockpuppet attacks.

With the recent COVID-19 outbreak, we have assisted to the development of new epidemic models or the application of existing methodologies to predict the virus spread and to analyze how the different lock-down strategies can effectively influence the epidemic diffusion. In this paper, we propose a novel machine learning based framework able to estimate the parameters of any epidemiological model, such as contact rates and recovery rates, based on static and dynamic features of places. In particular, we model mobility data through a graph series whose spatial and temporal features are investigated by combining Graph Convolutional Neural Networks (GCNs) and Long short-term memories (LSTMs) in order to infer the parameters of SIR and SIRD models. We evaluate the proposed approach using data related to the COVID-19 dynamics in Italy and we compare the forecasts of the trained model with available data about the epidemic spread.

The interest of Industry 4.0 in smart contracts and blockchain technologies is growing up day by day. Smart contracts have enabled new kinds of interactions whereby contractors can even fully automate processes they agree on. This technology is really appealing in Internet of Things (IoT) domain because smart devices generate events for software agents involved in a smart contract execution, making full automation possible. However, smart contracts have to comply with national and international laws and accountability of participant's actions. Soundness of a smart contract has to be verified in terms of law compliance. Here, we propose a model for verification and validation of law compliance of smart contracts in IoT environments. The main goal of this article is to propose a formal model (based on multiagent logic and ontological description of contracts) for validating law compliance of smart contracts and to determine potential responsibilities of failures.

Viral marketing is the modern version of the old “word-of-mouth” advertising, where companies choose a restricted number of persons, considered “influential,” recommending them products or services that will be in turn iteratively suggested. In this article, we propose cognitive models and algorithms for marketing applications through online social networks, considered as a graph database, and define the concept of influence graph leveraging particular user behavioral patterns, by querying the initial heterogeneous graph network. We also model the diffusion across the network, without any preliminary information, as a combinatorial multiarmed bandit problem, for the selection of most influential users. We have used the YELP social network as a case study for our approach, showing how it is possible to generate an influence graph considering several kinds of relevant paths (mainly considering reviews to the same firms) by which a user can influence other ones. Several experiments have been carried out and discussed, putting into evidence the effectiveness and efficacy of the proposed methods for influence maximization with respect to other approaches of state of the art.

The uncontrolled growth of fake news creation and dissemination we observed in recent years causes continuous threats to democracy, justice, and public trust. This problem has significantly driven the effort of both academia and industries for developing more accurate fake news detection strategies. Early detection of fake news is crucial, however the availability of information about news propagation is limited. Moreover, it has been shown that people tend to believe more fake news due to their features [10]. In this paper, we present our framework for fake news detection and we discuss in detail an approach based on deep learning that we implemented by using Google Bert features. Our experiments conducted on two well-known and widely used real-world datasets suggest that our method can outperform the state-of-the-art approaches and allows fake news accurate detection, even in the case of limited content information.

We demonstrate GraphDBLP, a tool to allow researchers for querying the DBLP bibliography as a graph. The DBLP source data were enriched with semantic similarity relationships computed using wordembeddings. A user can interact with the system either via a Web-based GUI or using a shell-interface, both provided with three parametric and pre-defined queries. GraphDBLP would represent a first graph-database instance of the computer scientist network, that can be improved through new relationships and properties on nodes at any time, and this is the main purpose of the tool, that is freely available on Github. To date, GraphDBLP contains 5+ million nodes and 24+ million relationship.

Recommender systems are increasingly playing an important role in our life, enabling users to find “what they need” within large data collections and supporting a variety of applications, from e-commerce to e-tourism. In this paper, we present a Big Data architecture supporting typical cultural heritage applications. On the top of querying, browsing, and analyzing cultural contents coming from distributed and heterogeneous repositories, we propose a novel user-centered recommendation strategy for cultural items suggestion. Despite centralizing the processing operations within the cloud, the vision of edge intelligence has been exploited by having a mobile app (Smart Search Museum) to perform semantic searches and machine-learning-based inference so as to be capable of suggesting museums, together with other items of interest, to users when they are visiting a city, exploiting jointly recommendation techniques and edge artificial intelligence facilities. Experimental results on accuracy and user satisfaction show the goodness of the proposed application.

The main objective of this work is the proposal of a decentralized data structure storing a large amount of data under the assumption that it is not possible or convenient to use a single workstation to host all data.The index is distributed over a computer network and the performance of the search, insert, delete operations are close to the traditional indices that use a single workstation.It is based on k-d trees and it is distributed across a network of "peers", where each one hosts a part of the tree and uses message passing for communication between peers.In particular, we propose a novel version of the k-nearest neighbour algorithm that starts the query in a randomly chosen peer and terminates the query as soon as possible.Preliminary experiments have demonstrated that in about 65% of cases it starts a query in a random peer that does not involve the peer containing the root of the tree and in the 98% of cases it terminates the query in a peer that does not contain the root of the tree.

Summary Nowadays, social networks provide users an interactive platform to create and share heterogeneous content for a lot of different purposes (eg, to comment events and facts, and express and share personal opinions on specific topics), allowing millions of individuals to create online profiles and share personal information with vast networks known and sometimes also unknown people. Knowledge about users, content, and relationships in a social network may be used for an adversary attack of some victims easily. Although a number of works have been done for data privacy preservation on relational data, they cannot be applied in social networks and in general for big data analytics. In this paper, we first propose a novel data model that integrates and combines information on users belonging to 1 or more heterogeneous online social networks, together with the content that is generated, shared, and used within the related environments, using an hypergraph data structure; then we implemented the most diffused centrality measures and also introduced a new centrality measure—based on the concept of “neighborhood” among users—that may be efficiently applied for a number of data privacy issues, such as lurkers and neighborhood attack prevention, especially in “interest‐based” social networks. Some experiments using the Yelp dataset are discussed.

Despite the great amount of research done in the Online Social Networks (OSNs) field, only few works have investigated the use of multimedia data in such realm. Instead, it is the authors' opinion that a novel data model that takes into account the intrinsic characteristics of multimedia may be of great help in managing Multimedia OSNs for providing more effective algorithms. In this paper, we describe a novel OSN data model that supports easy management of multimedia content in a unique framework, providing a more effective and efficient mechanism for data and information management in a variety of applications, especially for Influence Analysis aims.

Nowadays, the need for well-structured ontologies in the medical domain is rising, especially due to the significant support these ontologies bring to a number of groundbreaking applications, such as intelligent medical diagnosis system and decision-support systems. Indeed, the considerable production of clinical data belonging to restricted sub domains has stressed the need for efficient methodologies to automatically process enormous amounts of un-structured, domain specific information in order to make use of the knowledge these data provide. In this work, we propose a lexicon-grammar based methodology for efficient information extraction and retrieval on unstructured medical records in order to enrich a simple ontology descriptive of such a kind of documents. We describe the NLP methodology for extracting RDF triples from unstructured medical records, and show how an existing ontology built by a domain expert can be populated with the set of triples and then enriched through its linking to external resources.

In this paper, we present a framework supporting information retrieval over corpora of documents using an automatic sematic query expansion approach. The main idea is to expand the set of words used as query terms exploiting the notion of semantic similarity between the concepts related to the search terms. We leverage existing lexical resources and similarity metrics computed among terms to generate - by a proper mapping into a vectorial space - an index for the fast retrieval of a set of terms "semantically correlated" to a given query term. The vector of expanded terms is then exploited in the query stage to retrieve documents that are significantly related to specific combinations of the query terms. Preliminary experimental results concerning efficiency and effectiveness of the proposed approach are reported and discussed.

As well known, Semantic Web technologies make available a set of facilities that allow data to be shared and reused across applications. The last generation of Content Management System (CMS) can leverage such technologies to improve the content management task incorporating semantic annotations of the produced resources. Here, we present the benefits deriving from the application of semantic technologies in a CMS environment. To this goal, we collect preliminary results about the effectiveness of the integration of a semantic annotation engine within the Intrage Web Portal for content management purposes. The obtained results show that the approach is quite promising and encourage the current research.

Transaction management in different application contexts is still a challenging task. In this paper we propose a novel method in order to improve concurrency of a particular kind of transaction, known as long running transactions. Differently from other techniques presented in the literature, we design a sort of hybrid approach between optimistic and pessimistic concurrency models. From one hand, our basic idea consists in taking into account frequent disconnections or inactivity periods of a generic transaction during its life-cycle and, from the other one, we consider the semantics related to operations produced by transactions. Our solution avoid an indefinite or long resource locking due to disconnecting (or idle) transactions or a high rate of preventive aborts; eventually, a transaction semantic compatibility is exploited in order to increase the concurrency of reconcilable operations on the same resources. To these purposes, we have implemented a middleware with the aims of emulating a transactional scheduling, and several experiments have been carried out.

In this paper we present our development experience of a Semantic Content Management System able to handle and manage uniformly heterogeneous contents of different kinds (texts, video and images) considering the related semantics. To this aim, we exploit several Semantic Web technologies: RDF/OWL for data modeling and representation, SPARQL as querying language, Multimedia Information Extraction techniques, W3C standard models for creating taxonomies and resource annotation, vocabularies and microformats. We also follow the Best Practices and Issues for the Web of Data by reusing LOD information for the Entity annotation of the managed content, thus providing CMS with advanced capabilities such as contents’ authoring and tagging.

There are numerous applications (e.g., video surveillance, fraud detection, cybersecurity) in which we wish to identify unexplained sets of events. Most related past work has been domain-dependent (e.g., video surveillance, cybersecurity) and has focused on the valuable class of statistical anomalies in which statistically unusual events are considered. In contrast, suppose there is a set A of known activity models (both harmless and harmful) and a log L of time-stamped observations. We define a part L '⊆ L of the log to represent an unexplained situation when none of the known activity models can explain L ' with a score exceeding a user-specified threshold. We represent activities via probabilistic penalty graphs (PPGs) and show how a set of PPGs can be combined into one Super-PPG for which we define an index structure. Given a compute cluster of ( K + 1) nodes (one of which is a master node), we show how to split a Super-PPG into K subgraphs, each of which can be independently processed by a compute node. We provide algorithms for the individual compute nodes to ensure seamless handoffs that maximally leverage parallelism. PADUA is domain-independent and can be applied to many domains (perhaps with some specialization). We conducted detailed experiments with PADUA on two real-world datasets—the ITEA CANDELA video surveillance dataset and a network traffic dataset appropriate for cybersecurity applications. PADUA scales extremely well with the number of processors and significantly outperforms past work both in accuracy and time. Thus, PADUA represents the first parallel architecture and algorithm for identifying unexplained situations in observation data, offering both scalability and accuracy.

The extraordinary technological progress we have witnessed in recent years has made it possible to generate and exchange multimedia content at an unprecedented rate. As a consequence, massive collections of multimedia objects are now widely available to a large population of users. As the task of browsing such large collections could be daunting, Recommender Systems are being developed to assist users in finding items that match their needs and preferences. In this article, we present a novel approach to recommendation in multimedia browsing systems, based on modeling recommendation as a social choice problem. In social choice theory, a set of voters is called to rank a set of alternatives, and individual rankings are aggregated into a global ranking. In our formulation, the set of voters and the set of alternatives both coincide with the set of objects in the data collection. We first define what constitutes a choice in the browsing domain and then define a mechanism to aggregate individual choices into a global ranking. The result is a framework for computing customized recommendations by originally combining intrinsic features of multimedia objects, past behavior of individual users, and overall behavior of the entire community of users. Recommendations are ranked using an importance ranking algorithm that resembles the well-known PageRank strategy. Experiments conducted on a prototype of the proposed system confirm the effectiveness and efficiency of our approach.

Nowadays Augmented Reality (AR) is an emerging technology that promises significant results in a variety of applications. Despite the great advances in computer graphics and multimedia technologies, there is still a lack in the definition and building of a complete 3D object management system, in order to efficiently store, manage and retrieve such objects. In this paper a novel data model and a building and querying language for 3D objects are presented. This preliminary environment has been properly designed to be sufficiently powerful for including the functionalities and characteristics of modern 3D description languages, such as X3D and Collada, and at same time provides a formalism for optimization aims. Several examples from real Italian Cultural Heritage sites are provided and preliminary results are commented and discussed, showing the main advantages of the proposed system for both creation and information retrieval goals.

The definition of ontologies within the multimedia domain still remains a challenging task, due to the complexity of multimedia data and the related knowledge. In this paper, we present a novel framework (MOWIS) that aims at realizing a system for building Multimedia Ontologies from Web Information Sources. In particular, we propose: i) a multimedia ontology model that combines both low level descriptors and high level semantic concepts; ii) automatic construction of ontologies using the Flickr web services that provide images, tags, keywords and sometimes useful annotation describing both the image content and personal interesting information. Eventually, we describe an example of automatic ontology generation in a specific domain and present some preliminary experimental results.

The exchange of information between heterogeneous and distributed health information systems preserving the semantics is an important open issue for the health care sector. In this paper, we propose an approach that, exploiting the Semantic Web technologies, has the objective of allowing semantic interoperability among software agents for e-Health applications that preserves, not only the semantic of transmitted messages, but also the subjectivity of agent's world vision in the communication The proposed approach exploits the Semantic Triangle Model to differentiate the roles of referents (real world objects) and concepts (mental image or impression about a real object) in the communication process and ensures an effective semantic interoperability without the need of a unique shared conceptualization.

In this paper we propose a novel communication approach and a possible implementation of it that, exploiting the Semantic Web technologies, allows semantic interoperability among software agents in the Web, preserving not only the semantics but also the subjectivity of the agent's world vision in the communication. Such an approach takes advantage of a particular communication model called Semantic Triangle in which communication agents share the referents (real world objects) and not the concepts (mental image or impression of a real object from the sender agents point of view), thus ensuring an effective semantic interoperability in the information exchange process. The proposed approach has been submitted to an experiment that involved an instantiation of the communication process based on semantic machines and showed the approach feasibility for the semantic information exchange and its effectiveness.

Recommender systems in e-commerce applications help consumers with information useful to decide which products to purchase, suggesting products, services, and information items to potential consumers. Nowadays recommender systems interfaces are more oriented to technical people than to normal consumer, that are really not necessarily expert of statistic, scores and so on. In this paper we propose to join the capabilities of relevance feedback with recommendation strategies in a more useful architecture based on 3D navigation systems. A general framework is described, together with novel techniques oriented to an effective human-computer interaction. An example of the proposed system is also discussed.

E-government processes are dedicated to the improvement of the efficiency, inexpensiveness and accessibility of public administration services: dematerialization activities, introduced to manage bureaucratic digital documents in a proper way, are among the main tasks of the e-government works. In this paper we present a novel RDF model of digital documents for improving the dematerialization effectiveness, that constitutes the starting point of an information system able to manage documental streams in the most efficient way. Such model takes into account the important need that is required in several e-government applications which, depending on authorities or final users or time, provides different representations of the same multimedia contents.

In this paper, we present a new vision of multimedia recommender systems based on an a novel paradigm that combines both analysis of user behavior and semantic descriptors of multimedia objects. In particular, we model recommendation as a two step process: First we propose a model of user behavior, based on a semantic network that captures the necessary knowledge of the domain of interest; second we propose a recommendation strategy based on both this kind of a-priori knowledge and on the current usage patterns, considering the actual and complex structure of multimedia data. We have implemented a prototype that supports our proposed recommendation strategy; eventually a real use of our system based on a 3D interface is presented and discussed.

In the classical theory of social choice, a set of voters is called to rank a set of alternatives and a social ranking of the alternatives is generated. In this paper, we model recommendation in the context of browsing systems as a social choice problem, where the set of voters and the set of alternatives both coincide with the set of objects in the data collection. We then propose an importance ranking method that strongly resembles the well known PageRank ranking system, and takes into account both the browsing behavior of the users and the intrinsic features of the objects in the collection. We apply the proposed approach in the context of multimedia browsing systems and show that it can generate effective recommendations and can scale well for large data collections.

There is now a growing need to identify various kinds of activities that occur in videos. In this paper, we first present a logical language called Probabilistic Activity Description Language (PADL) in which users can specify activities of interest. We then develop a probabilistic framework which assigns to any subvideo of a given video sequence a probability that the subvideo contains the given activity, and we finally develop two fast algorithms to detect activities within this framework. OffPad finds all minimal segments of a video that contain a given activity with a probability exceeding a given threshold. In contrast, the OnPad algorithm examines a video during playout (rather than afterwards as OffPad does) and computes the probability that a given activity is occurring (even if the activity is only partially complete). Our prototype Probabilistic Activity Detection System (PADS) implements the framework and the two algorithms, building on top of existing image processing algorithms. We have conducted detailed experiments and compared our approach to four different approaches presented in the literature. We show that-for complex activity definitions-our approach outperforms all the other approaches.

E-government activities have the aim to optimize the work of the governmental offices and offer, to citizens and businesses, effective, accessible and efficient services. Dematerialization, which is among the main activities of e-government, is devoted to optimize information communication, in terms of consumed time and resources, as well as processing documents. In this work we propose a new model for multimedia document, suitable for e-government activities, and an information system able to integrate and process different multimedia data type (as images, text, graphic objects, audio, video, composite multimedia, etc.) providing facilities for indexing, storage, retrieval, control of multimedia data, together with long term preservation strategies.

In this paper, we propose an architecture that, exploiting the Semantic Web technologies, has the objective of allowing semantic interoperability among software agents in the Web. Such an architecture takes advantage by the Semantic Triangle model in which communication agents share the referents (real world objects) and not the references (mental image of a real object of the sender agent), thus ensuring an effective semantic interoperability in the information exchange process. We have carried a case study in order to assess the appropriateness and the feasibility of the process for the semantic information exchange by realizing and testing an instantiation of the related architecture.

Traditional image classification techniques are based on the analysis of low-level visual features or on textual information. In this paper, we describe a novel solution which tries to improve image analysis and processing algorithms by incorporating keywords and textual annotation produced by humans in a folksonomy environment, namely Flickr. The obtained experimental results demonstrate that the proposed categorization process achieves quite good performances in terms of efficiency and effectiveness.

Recognition of human activities in restricted settings such as airports, parking lots and banks is of significant interest in security and automated surveillance systems. In such settings, data is usually in the form of surveillance videos with wide variation in quality and granularity. Interpretation and identification of human activities requires an activity model that a) is rich enough to handle complex multi-agent interactions, b) is robust to uncertainty in low-level processing and c) can handle ambiguities in the unfolding of activities. We present a computational framework for human activity representation based on Petri nets. We propose an extension-Probabilistic Petri Nets (PPN)-and show how this model is well suited to address each of the above requirements in a wide variety of settings. We then focus on answering two types of questions: (i) what are the minimal sub-videos in which a given activity is identified with a probability above a certain threshold and (ii) for a given video, which activity from a given set occurred with the highest probability? We provide the PPN-MPS algorithm for the first problem, as well as two different algorithms (naive PPN-MPA and PPN-MPA) to solve the second. Our experimental results on a dataset consisting of bank surveillance videos and an unconstrained TSA tarmac surveillance dataset show that our algorithms are both fast and provide high quality results.

Recognition of human activities in restricted settings such as airports, parking lots and banks is of significant interest in security and automated surveillance systems. In such settings, data is usually in the form of surveillance videos with wide variation in quality and granularity. Interpretation and identification of human activities requires an activity model that a) is rich enough to handle complex multi-agent interactions, b) is robust to uncertainty in low-level processing and c) can handle ambiguities in the unfolding of activities. We present a computational framework for human activity representation based on Petri nets. We propose an extension—Probabilistic Petri Nets (PPN)—and show how this model is well suited to address each of the above requirements in a wide variety of settings. We then focus on answering two types of questions: (i) what are the minimal sub-videos in which a given activity is identified with a probability above a certain threshold and (ii) for a given video, which activity from a given set occurred with the highest probability? We provide the PPN-MPS algorithm for the first problem, as well as two different algorithms (naive PPN-MPA and PPN-MPA) to solve the second. Our experimental results on a dataset consisting of bank surveillance videos and an unconstrained TSA tarmac surveillance dataset show that our algorithms are both fast and provide high quality results.

The aim of the IDEA (Instructional DEsigner Applications) is to enhance contents production process making available to the content owners very simple and useful authoring tools for the digitalization, organization and classification of their knowledge. The results of a such process can be seen as prototypes of multimedia courses, obtained with reduced costs and times, that can be directly imported into a generic Learning Management Systems (LMS) or used as startpoint in generating more complex multimedia products. .

We view scenes in the real world by moving our eyes three to four times each second and integrating information across subsequent fixations (foveation points). By taking advantage of this fact, in this paper we propose an original approach to partitioning of a video into shots based on a foveated representation of the video. More precisely, the shot-change detection method is related to the computation, at each time instant, of a consistency measure of the fixation sequences generated by an ideal observer looking at the video. The proposed scheme aims at detecting both abrupt and gradual transitions between shots using a single technique, rather than a set of dedicated methods. Results on videos of various content types are reported and validate the proposed approach.