Nowadays, content proliferation makes protecting intellectual property vital. Digital watermarking has emerged as a key tool in the protection of digital content, striking a balance between safeguarding content and allowing its distribution without hindrances. Of all types of content, text is the hardest to watermark because of the low embeddability of the medium. This article presents one new structural watermarking technique and various combinations of it with others we previously proposed as a way to improve embeddability and to offer greater robustness to attacks. The efficacy of these new techniques has been validated against a novel suite of attacks designed to emulate the operational methodologies a human attacker might utilize to compromise protected content. We consequently propose this attack suite as a new, standardized benchmark for the rigorous assessment of future techniques. Robustness is evaluated across different payload sizes at varying attack magnitudes, starting from 10

Background/Objectives: In the Emergency Department (ED), non-critical patients are classified as Triage Level (TL) 3 on arrival if they are assessed as having a high-level of complexity (HLC), or as TL4-5 if they are assessed as having a low-to-mild level of complexity (LLC). These levels are based on the estimated resources needed. This study aimed to identify the characteristics associated with an HLC or LLC by considering a group of variables from the presentation profile (PP) and clinical diagnostic workload (CDW), assessing ex post whether the assignment of complexity levels based on a priori estimation of the number of resources needed can be considered adequate. Materials and Methods: This retrospective multicentre study involved four first-level EDs and included patients between 2023 and 2024. Outcome Measures: The variables tested in a logistic model were those of the PP (age, sex, chief complaint, National Early Warning Score (NEWS), Numeric Rating Scale (NRS) and those of the CDW (diagnostic tests, interventions and therapy, assistance, and ED length of stay). Results: Of the 335,507 subjects considered, the average age was 59 years (interquartile range [IQR], 25), with 43.3% of cases classified as TL3. An NRS ≥ 7, ECG, urgent laboratory tests, NEWS > 6, need for a stretcher, and male gender were associated with TL3, whereas obstetric-gynaecological complaints, environmental complaints, skin-presenting complaints, and intramuscular therapy were associated with TL4-5. Conclusions: In non-critical patients a defined group of features were associated with different levels of complexity, going beyond the standard criterion based on the resources needed. These results could help clinicians improve the appropriateness of ED care pathways.

Background and Objective: Speech has emerged as a low-cost and non-invasive digital biomarker with considerable potential for cognitive impairment detection. However, limited labeled data and cross-dataset variability remain major challenges for robust speech-based screening systems. Methods: We developed a segment-level representation learning framework for speech-based cognitive impairment detection. Speech recordings were divided into short segments and converted into spectrogram representations. To improve robustness under limited-data conditions, offline and online augmentation strategies were combined with autoencoder-based representation learning and contrastive objectives to enhance discriminative latent representations. Results: Experiments conducted on four independent Mandarin Chinese speech datasets demonstrated stable and competitive performance in both binary and three-class classification tasks, with particularly notable improvements in the clinically challenging three-class setting. Ablation studies further supported the effectiveness of the proposed framework. Conclusions: The findings suggest that segment-level speech representation learning may provide a scalable and practical approach for cognitive impairment screening in resource-constrained clinical settings.

The task of authorship attribution is a classification problem where a given text is analyzed by a machine learning algorithm with the goal of verifying its author. This task has great application in the world of forensic analysis and intellectual property problems. In this work, we propose to use Support Vector Machines paired with Term Frequency - Inverse Term Frequency to solve authorship attribution in cases where the author is part of a predetermined set of authors, called a closed set. This method could also be applied to large language models in order to find out if those have a "style" of writing. In the end, we show how we compared our results with related works and how we managed to improve the results on a well known benchmark dataset by 6.4 %.

The increasingly large amount of available biomedical literature is making it difficult to gather and synthesise all the necessary information. Moreover, this domain-specific task demands a high level of reliability in the generated text and concepts. Pre-trained large language models have recently shown promising results. Given the specific requirements of biomedical text summarisation, our evaluation focuses on extractive models, prioritising the accuracy of the generated text. In this article, we evaluate the capabilities of 18 general-domain and biomedical pre-trained language models in various configurations on the biomedical extractive summarisation task using one single and two multi-document datasets consisting of 33,000, 5,000 and 470,000 PubMed articles, respectively. We performed the comparison using several well-known metrics, namely ROUGE-1, ROUGE-2, ROUGE-L, BERTScore, BLEU and METEOR. The main contribution of this work lies in providing a detailed performance analysis, highlighting the differences between general-domain and biomedical models, and identifying key factors that influence model performance in extractive summarisation tasks within the biomedical domain. Experimental results show that biomedical models tend to result in higher recall, while general-domain models produce higher precision, while general-domain models produce higher precision. This corresponds to more expressive summaries for biomedical models and shorter summaries for general-domain models.

The rapid proliferation of online information disorder poses a significant societal challenge. This phenomenon has been further exacerbated by the pervasive influence of social media, affecting a broad range of domains. Addressing the spread of information disorder through manual approaches (e.g., human fact-checking) is impractical due to the vast volume of textual content daily generated, often facilitated by disruptive generative technologies. Similarly, the implementation of automated tools presents considerable obstacles, primarily due to the inherent ambiguity and complexity of natural language. This paper introduces the CERVANTES platform: an innovative application of blockchain technology integrated with text watermarking techniques. It is conceived to support news producers and enhance awareness among readers engaging with content across various social media platforms. CERVANTES is a multiplatform and language-independent solution that addresses online information disorder from an innovative perspective. It allows news producers to automatically embed a unique watermark within the content they create, ensuring the integrity and authenticity of the text, and safeguarding it against manipulation and misattribution. Furthermore, the association between the watermark and the corresponding news item is securely recorded on the blockchain, mitigating the risk of manipulation that might result from centralised management. We conduct an extended evaluation on twelve different social media platforms with a cohort of twenty users, observing ease of use and a high degree of satisfaction.

Limited hospital resources may prolong patient stays in the Emergency Department (ED), potentially affecting clinical outcomes. This paper investigates the link between overnight Emergency Department (ED) stays and in-hospital mortality, focusing on comorbidity extraction from clinical records. In Italian healthcare records, comorbidities are typically documented using abbreviations and non-standard clinical slang in unstructured free-text fields. We evaluated two approaches for comorbidity extraction: a rule-based method and a Large Language Model approach. Both were assessed against a dataset of 200 clinical records manually annotated by emergency medical staff. This first result showed that the rule-based strategy outperformed Large Language Models in terms of recall, F1-score, consistency, and reliability. Then, to assess the impact of overnight stays on in-hospital mortality and to identify the most significant predictors, 126,696 ED admissions at the Romagna Local Health Agency in Forli, Italy, between 2017 and 2022 were analysed using several models, with particular emphasis on interpretability. Comorbidity burden, diagnosis severity, age, and infectious, respiratory, and circulatory diseases emerged as the most influential factors.

This paper presents a complete explainable system that interprets a set of data, abstracts the underlying features and describes them in a natural language of choice. The system relies on two crucial stages: (i) identifying emerging properties from data and transforming them into abstract concepts, and (ii) converting these concepts into natural language. Despite the impressive natural language generation capabilities demonstrated by Large Language Models, their statistical nature and the intricacy of their internal mechanism still force us to employ these techniques as black boxes, forgoing trustworthiness. Developing an explainable pipeline for data interpretation would allow facilitating its use in safety-critical environments like processing medical information and allowing non-experts and visually impaired people to access narrated information. To this end, we believe that the fields of knowledge representation and automated reasoning research could present a valid alternative. Expanding on prior research that tackled the first stage (i), we focus on the second stage, named Concept2Text. Being explainable, data translation is easily modeled through logic-based rules, once again emphasizing the role of declarative programming in achieving AI explainability. This paper explores a Prolog/CLP-based rewriting system to interpret concepts-articulated in terms of classes and relations, plus common knowledge-derived from a generic ontology, generating natural language text. Its main features include hierarchical tree rewritings, modular multilingual generation, support for equivalent variants across semantic, grammar, and lexical levels, and a transparent rule-based system. We outline the architecture and demonstrate its flexibility through some examples capable of generating numerous diverse and equivalent rewritings based on the input concept.

Information disorder has become a major societal challenge, impacting public discourse and democracy. This phenomenon has been exacerbated by the spread of social media platforms, affecting various areas, ranging from national elections to public health. Addressing fake news through a manual approach (e.g., human fact-checking) is unfeasible due to the rapid production of textual content. At the same time, applying automatic tools is equally challenging, primarily due to the ambiguity of natural language. In this paper, we addressed online information disorder from a different perspective by proposing a platform that supports trustworthy and reputable news producers and enhances awareness among readers across various social media. Specifically, the proposed platform enables news producers to automatically embed a unique watermark in the text they create, ensuring that the news cannot be manipulated or misattributed. The watermarking is embedded in a fine-grained way, allowing even small extracts of the news to be shared while preserving traceability. Additionally, the association between the watermark and the news item is recorded in a distributed ledger, preventing further manipulation that could arise from centralised management. The aim is to enable readers to make more informed decisions about the content they encounter, even when engaging with excerpts of the original document, minimising reliance on external fact-checking organisations.

Background Overcrowding, prolonged stays and invasive interventions could increase the risk of healthcare-associated infections (HAIs) in Emergency Departments (ED). Aim of study was to investigate whether the risk of developing a HAI can be estimated in patients at entry on the basis of ED visit data, and whether they are associated with poorer outcome. Methods This retrospective single centre study included subjects who required urgent hospitalisation following ED visit between 2017 and 2022. Main outcome measures considered were the occurrence of late HAIs and the cumulative survival rate at 30 days. The key predictive features tested in a logistic model were age, sex, vital parameters as measured by the National Early Warning Score (NEWS), priority levels upon entry, comorbidities by the Charlson Comorbidity Index (CCI), trauma related diseases, main diagnosis and ED length of stay. Results In 2,556 (8,9%) out of 28,803 hospitalised patients aged 73 [17] years (mean [SD]) a diagnosis of HAI was recorded during hospital stays. In order of frequency, HAIs occurred in bloodstream (4.7%), in urinary (3.4%), respiratory (2.9%), gastrointestinal (0.4%) tract, or in surgical (0.3%) and skin and soft tissue (0.05%) sites. Main features selected by the logistic model in the prediction of HAI were infectious and parasitic diseases, CCI > 4, genitourinary system diseases, CCI 3 to 4, COVID period, priority level at arrival in ED. In-hospital cumulative survival rate in HAI group was reduced, namely for subjects with pneumonia and bloodstream infections. Conclusions A group of key characteristics in subjects visiting the ED can predict the onset of nosocomial infections that negatively affect survival, particularly for patients who develop pneumonia or bloodstream infections.

The concept of luxury, considering it a rare and exclusive attribute, is evolving due to technological advances and the increasing influence of consumers in the market. Luxury cars have always symbolized wealth, social status, and sophistication. Recently, as technology progresses, the ability and interest to gather, store, and analyze data from these elegant vehicles has also increased. In recent years, the analysis of luxury car data has emerged as a significant area of research, highlighting researchers’ exploration of various aspects that may differentiate luxury cars from ordinary ones. For instance, researchers study factors such as economic impact, technological advancements, customer preferences and demographics, environmental implications, brand reputation, security, and performance. Although the percentage of individuals purchasing luxury cars is lower than that of ordinary cars, the significance of analyzing luxury car data lies in its impact on various aspects of the automotive industry and society. This literature review aims to provide an overview of the current state of the art in luxury car data analysis.

Due to urbanisation and increased environmental awareness, many developed countries have put much effort into promoting sustainable mobility modes. Cycling is one option to replace motorised transport in cities. However, this shift often requires the transformation of the urban environment to be more suitable for cyclists. In this paper, we study the spatial and temporal patterns of cyclists and the influence of cycling infrastructure on their route choices using OpenStreetMap street tag information. A dataset of 290,117 unique trips collected over six months in Bologna, Italy, was used to gain insight into cyclist behaviour and preferences in an urban environment. The bicycle in Bologna is used as a means of transportation mainly for short-and medium-distance trips. The analysis of this dataset revealed the routine of cyclists in different periods: peak hours in the workday morning and evening and periods with lower activities during weekends. Generally, cyclists prefer streets with good cycling infrastructure and avoid streets with heavy traffic. In the suburban area, cyclists prefer bigger roads which lead towards the city centre, while in the city centre preference is on roads which avoid narrow streets in the old town. The obtained results may help to grasp the reasons behind specific route preferences, and they can be used to find opportunities to improve cycling infrastructure for a better and safer urban environment.

A bankruptcy prediction model is essential for stakeholders to avoid losses and market imbalances arising from the misallocation of resources. In the literature, several methods have been proposed for bankruptcy prediction. However, conventional statistical techniques and financial ratios have limitations in correctly identifying financially distressed firms in a reasonable time frame. In this study, we compare several machine learning approaches to define a bankruptcy prediction model using raw accounting data from financial statements. In particular, we used a dataset of 1'826'157 financial statements from 532'255 active and 76'464 bankrupt Italian firms (i.e., the last three financial statements for each firm), considering both financial ratios and the raw accounting data as input variables. To overcome the constraints of imbalance between the two classes of bankrupt and active firms, we used rebalancing techniques. We found that Random Forest is the best-performing model, exhibiting an accuracy of 98% which is an increase of 13.95% compared to the use of the financial ratios. Moreover, relative variable importance analysis shows that equity/total debts ratio and short-term debts are the most essential bankrupt predictors.

Many works have shown associations between social determinants of health (SDoH) -the social circumstances in which people live-and health-related outcomes.However, the lack of SDoH data increases the challenges in measuring and understanding their effect on people's health and health systems.In this paper, we present GASTon, a system for the indexing, annotation, and graph-based rendering of PubMed information to enable the search and retrieval of SDoHs in scientific literature.Our work provides a way to associate specific concepts with peer-reviewed articles to simplify the search for social factors.It builds a knowledge graph based on PubMed publications and associates them with concepts extracted from the Unified Medical Language System (UMLS) Metathesaurus.GASTon allows a full-text search and graph-based navigation and supports an overview of the concepts and related publications.Moreover, the architecture allows scale-up thanks to its containerized nature and parallelization capabilities.The system is open-source under the Apache V2 license.

As a result of an increasing elderly population, the number of people with age-related diseases is increasing worldwide. Alzheimer's disease is thus becoming an emergency health and social problem. Neuropsychological evaluation and biomarker identification represent the two main approaches to identifying subjects with Alzheimer's. In this paper, we propose a web application designed to be sensitive to the cognitive changes distinctive of the early Mild Cognitive Impairment, which is a condition in which someone experiences minor cognitive problems, and the preclinical phase of Alzheimer's disease. The application is conceived to be self-administered in a comfortable and non-stressful environment. It was designed to be quick to administer, automatic to score, and able to preserve privacy because of the highly sensitive data collected. The preliminary evaluation of the application was done by enrolling 518 subjects characterised by several risk factors and the presence of a family history, which underwent standard neuropsychological screening.

There is a need to persuade public and private entities to share their currently unexposed bio-data banks by preserving ownership and secrecy. The reason is to make available results that can be obtained by massively exploiting the content of such data by modern machine learning approaches. Digital catalogues of data collections are being provided. However, they are not developed to protect private content that may be shared according to privileges assigned by the owners. Here, we present BIOCHAIN, a data-sharing module which will be the basis for a computational platform aimed at performing federated data analysis. The platform is intended to be used by a consortium of private and public institutions in the field of microbiology. BIOCHAIN makes use of blockchain technology to guarantee fairness among entities of the consortium by allowing them to securely share their data.

Studies have shown that data retrieval and visualization tools can help health professionals to improve their understanding and communication with patients, their relationship with stakeholders, and their decision-making process. However, not many efforts have been made in this direction. In this paper, we present a prototype system for the indexing, annotation, and visualization of the PubMed knowledge base to enable the search and retrieval of health-related evidence. The proposed tool builds and keeps updated an enriched graph based on PubMed articles associating them with concepts extracted from the Unified Medical Language System (UMLS) Metathesaurus. Moreover, it allows a full-text search and graph-based navigation and supports an overview of concepts and related publications. The proposed architecture enables scale-up thanks to its containerized nature and parallelization capabilities. The code is open-source under the Apache V2 license.

BACKGROUND:The long-term results of web-based behavioural intervention in non-alcoholic fatty liver disease (NAFLD) have not been described in patients followed in specialised centres. AIMS:To analyse the long-term effectiveness of web education compared with the results achieved by a group-based behavioural intervention in the same years 2012-2014. METHODS:We followed 679 patients with NAFLD (web-based, n = 290; group-based, n = 389) for 5 years. Weight loss ≥10% was the primary outcome; secondary outcomes were attrition, changes in liver enzymes and in biomarkers of steatosis (Fatty liver Index) and fibrosis (Fibrosis-4 index). RESULTS:The cohorts differed in age, education, working status and presence of diabetes. Attrition was higher in the web-based cohort (hazard ratio: 1.53; 95% CI: 1.24-1.88), but not different after adjustment for confounders. Among patients in active follow-up, 50% lost ≥5% of initial body weight and 19% lost ≥10%, without difference between cohorts. Alanine aminotransferase levels fell to within the normal range in 51% and 45% of web- and group-based cohorts, respectively. Fatty Liver Index declined progressively and, by year 5, it ruled out steatosis in 4.8%, whereas 24.9% were in the indeterminate range. Fibrosis-4 index increased in both cohorts, driven by age, but the prevalence of cases ruling-in advanced fibrosis remained very low (around 1%). Improvements in the class of both surrogate biomarkers were associated with ≥5% weight loss. CONCLUSIONS:Although burdened by attrition, web-based behavioural intervention is feasible and effective in NAFLD, expanding the cohort involved in behavioural programs and reducing the risk of progressive disease.