Oral communication is a crucial skill in modern society. Nevertheless, it requires sustained practice and constructive feedback. Consequently, several studies have explored the development of oral communication trainers powered by Artificial Intelligence (AI). However, what characterizes expert speech remains unclear, especially given the need to adapt speech to contextual factors. In instructional environments, the speaker’s communication proficiency is a key determinant of audience learning outcomes. For this reason, we have analyzed 1250 speeches from five types of instructional discourses: in-person college classes ( Lectures ), online learning lessons ( Online Courses ), instructional animations ( Animated Lessons ), supplementary materials for school and high school ( Supplementary Lessons ), and public presentations ( Public Talks ). We extracted 16 speech metrics, including six additional multiple-participant metrics for Lectures . We obtained 250 videos of each discourse type, ensuring a minimum length of five minutes. Our analysis revealed expert values for each speech metric and showed how speech metrics vary across discourse types. We also developed an AI speech classifier that achieved an F1 score of 0.78. The model struggled to identify Online Courses , which is consistent with the Uniform Manifold Approximation and Projection analysis, showing that Online Courses are closely interjected with the speech of other instructional discourses. Furthermore, we identified distinct speech profiles in Lectures, Public Talks , and Online Courses , highlighting variations in speaking styles. This research provides valuable insights into expert speech in instructional discourses by offering reference values that can help speakers refine their delivery and support researchers in developing more effective speech training systems.

Abstract The existence of malicious software (malware) represents a potential threat to users who connect to a large set of services provided by multiple providers. Such malware is capable of stealing, spying on, encrypting data from users, and spreading, provoking impacts that are beyond a single citizen’s device and reaching critical information systems. To detect malware families, Machine Learning and Deep Learning techniques have been employed recently, demonstrating promising results. However, these techniques lack in detecting more advanced malware that employs obfuscation techniques. In this paper, we present DEFENDIFY, a novel framework, empowered by Computer Vision, Deep Learning, and Transfer Learning techniques, that is able to detect completely obfuscated malware with high performance in terms of accuracy and computational consumption. DEFENDIFY comprises three modules: Dataset Creation, Binary Obfuscation, and Model Generation. These modules work together to detect both obfuscated and non-obfuscated malware. The core module, i.e., the Model Generation, employs an entropy tester that determines whether a sample is obfuscated or not. Then, a Deep Learning model powered by Transfer Learning is employed to determine if it is malware or goodware. We validated our framework using real data gathered from malware repositories and legitimate software. The proposed framework was configured to test four Convolutional Neural Network architectures: ResNet18, ResNet34, EfficientNetB3, and EfficientNetV2S. Among them, the ResNet18 architecture obtained the best performance in detecting both non-obfuscated and obfuscated samples with an F1-score of 99.34% and 97.5%, respectively.

ABSTRACT The quality of education hinges on the proficiency and training of educators. Due to the importance of teacher training, the innovative platform Teacher Moments creates simulated classroom scenarios. In this scenario‐based learning, confusion is an important indicator to detect users who struggle with the simulations. Through Teacher Moments, we gathered 7975 audio recording responses from participants who self‐labelled their recordings according to whether they sounded confused. Our dataset stands out for its size, for not including actor‐generated audio, and for measuring confusion, a neglected emotion in artificial intelligence (AI). Our experiments tested unimodal approaches and feature‐level, model‐level and decision‐level fusion. Feature‐level fusion demonstrated superior performance to unimodal methods, achieving a balanced accuracy of 0.6607 on the test set. This outcome highlights the necessity for further investigation in the overlooked area of confusion detection, particularly employing realistic datasets like the one used in this study and exploring new methods. Beyond teacher training, the insights of this research also extend to other directions, such as other professionals making critical decisions, user interface design or adaptive learning systems.

Abstract In our generation, there is an undoubted rise in the use of social media and specifically photo and video sharing platforms. These sites have proved their ability to yield rich data sets through the users’ interaction which can be used to perform a data‐driven evaluation of capabilities. Nevertheless, this study reveals the lack of suitable data sets in photo and video sharing platforms and evaluation processes across them. In this way, our first contribution is the creation of one of the largest labelled data sets in Flickr with the multimodal data which has been open sourced as part of this contribution. It incorporates multimodal data, combining information from various sources such as user profiles, photo metadata, and crowdsourced features. Predicated on these data, we explored machine learning models and concluded that it is feasible to properly predict whether a user is a professional photographer or not based on self‐reported occupation labels and several feature representations out of the user, photo and crowdsourced sets. We also examined the relationship between the aesthetics and technical quality of a picture and the social activity of that picture. Finally, we depicted which characteristics differentiate professional photographers from non‐professionals. As far as we know, the results presented in this work represent an important novelty for identifying expertise in the domain of photography, which researchers from various domains can utilise for related applications.

Managing stress is a crucial soft skill that affects cognitive performance and health. Stress detection through biometrics can be used to improve and evaluate stress management. However, measuring the effects of stress on biometrics and executive functions is difficult and dependent on the individual. Despite these challenges, this paper presents a case study that collects a comprehensive multimodal dataset with two stress metrics, four biometric signals, and twenty-two executive function metrics from Game-based Assessment (GBA) trace data specifically designed for this purpose. The experiments suggest that biometrics, especially the heart rate and skin temperature, are effective predictors of stress. Additionally, noteworthy correlations were observed between heart rate and certain executive function variables. The levels of GBA that measured shifting and processing speed showed a higher heart rate than the response inhibition levels. This case study, together with the developed stress detectors, enables the detection of persons who struggle to manage stress and measure their executive function performance under stressful situations.

The dark web has become notorious for its association with illicit activities and there is a growing need for systems to automate the monitoring of this space. This paper proposes an end-to-end scalable architecture for the early identification of new Tor sites and the daily analysis of their content. The solution is built using an Open Source Big Data stack for data serving with Kubernetes, Kafka, Kubeflow, and MinIO, continuously discovering onion addresses in different sources (threat intelligence, code repositories, web-Tor gateways, and Tor repositories), downloading the HTML from Tor and deduplicating the content using MinHash LSH, and categorizing with the BERTopic modeling (SBERT embedding, UMAP dimensionality reduction, HDBSCAN document clustering and c-TF-IDF topic keywords). In 93 days, the system identified 80,049 onion services and characterized 90% of them, addressing the challenge of Tor volatility. A disproportionate amount of repeated content is found, with only 6.1% unique sites. From the HTML files of the dark sites, 31 different low-topics are extracted, manually labeled, and grouped into 11 high-level topics. The five most popular included sexual and violent content, repositories, search engines, carding, cryptocurrencies, and marketplaces. During the experiments, we identified 14 sites with 13,946 clones that shared a suspiciously similar mirroring rate per day, suggesting an extensive common phishing network. Among the related works, this study is the most representative characterization of onion services based on topics to date.

The rapid technological evolution of the last years has motivated students to develop capabilities that will prepare them for an unknown future in the 21st century. In this context, many teachers intend to optimise the learning process, making it more dynamic and exciting through the introduction of gamification. Thus, this article focuses on a data-driven assessment of geometry competencies, which are essential for developing problem-solving and higher-order thinking skills. Our main goal is to adapt, evaluate and compare Bayesian Knowledge Tracing (BKT), Performance Factor Analysis (PFA), Elo, and Deep Knowledge Tracing (DKT) algorithms applied to the data of a geometry game named Shadowspect, in order to predict students’ performance by means of several classifier metrics. We analysed two algorithmic configurations, with and without prioritisation of Knowledge Components (KCs) – the skills needed to complete a puzzle successfully, and we found Elo to be the algorithm with the best prediction power with the ability to model the real knowledge of students. However, the best results are achieved without KCs because it is a challenging task to differentiate between KCs effectively in game environments. Our results prove that the above-mentioned algorithms can be applied in formal education to improve teaching, learning, and organisational efficiency.

Abstract The Tor network is known for its opaque characteristics and involvement in illicit activities, motivating to shed light on the exposure, lifetime, and functionalities of onion services. This study focuses on the appearance of Tor links in online advertising and monitors the connectivity status and protocols of the collected onion domains through the Tor network over 105 days. Out of 54,602 onion addresses gathered, it was found that 38% of Tor links were advertised only once, 43% between two and five times, and 19% more than five times. Furthermore, 50% of the addresses were exclusively advertised on the surface web, 6% on the dark web, and 44% on both portions. The temporal analysis revealed that 67% of the addresses were predominantly active, 7% were intermittent, and 26% were mostly inactive. The study examined fifteen protocols used by onion services, concluding that 94% employed a single protocol, while 6% utilized between two and eight protocols. Among active sites, HTTP was present in 99.75% of cases, followed by SSH (4.95%) and HTTPS (0.64%). Additionally, onion services without web services often deploy cryptocurrency or instant messaging servers. This study offers a comprehensive and current understanding of the dark web, surpassing previous research in its scope.

During the past decade, there has been growing interest among researchers in informal learning at scale, particularly in the area of expert finding. These platforms have played a fundamental role in facilitating informal learning at scale, by providing access to diverse expertise and knowledge resources that might not otherwise be available to learners. Based on the encountered gaps in expert identification in Question & Answer (Q&A) portals, we inspect the feasibility of identifying data science experts in Reddit using the activity behaviour of every user, including Natural Language Processing (NLP), crowdsourced and user features sets. We also examine the impact of using only expert and non-expert classes versus three classes additionally including the out-of-scope class. Our findings can be used for distinguishing different types of users in Reddit, creating a recommendation system, identifying unreliable users or social bots in the early stage and reducing their influence.

Exploring the Tor network requires acquiring onion addresses, which are crucial for accessing anonymous websites. However, the Tor protocol presents a challenge, as it lacks a standard method for finding these complex links composed of either 16 or 56 base32-coded characters and featuring the unique “.onion” top-level domain. This study delves into the existing literature analyzing onion services and categorizes the various strategies employed to gather their addresses. The success of each approach is measured by the number of addresses obtained, while the relevancy of the work is evaluated by comparing the number of services uncovered to Tor’s official count. The results indicate that the most used techniques are Tor crawling and repositories, whereas the most effective methods are relay injection, repositories, and Tor crawling. This paper also estimates the representativeness of literature collections, revealing that most past works explored a small portion of the Tor network. The study also uncovers the limitations of onion gathering and sheds light on the challenges for future research to provide more representative datasets for dark web exploration.

Social networks were designed to connect people online but have also been exploited to launch influence operations for manipulating society. The deployment of social bots has proven to be one of the most effective enablers to polarize and destabilize platforms. While automatic tools have been developed for their detection, the way to characterize these accounts and measure their impact is heterogeneous in the literature. In this work, we select metrics and algorithms from existing efforts to ensemble a data-driven methodology to profile groups of users and bots of Twitter from seven perspectives. We apply the framework to a dataset of Twitter retweets before the 10 November 2019 Spanish elections to characterize potential interferences. In this case study, Likely Bots (fully automated accounts) and Likely Semi-Bots (partially automated accounts) interacted with the same tendencies as Likely Humans (non-automated users), generating similar virality (information cascades) over time and without compromising the network connectivity. However, Likely Bots particularly stood out as close, visible, and reachable to other users. Likely Semi-Bots attracted particular attention, created proportionally more retweets, and were placed in strategically key positions in the core of the network. Results suggest that semi-automated accounts would be more threatening than fully automated ones.

Cyberattacks targeting modern network infrastructures are increasing in number and impact. This growing phenomenon emphasizes the central role of cybersecurity and, in particular, the reaction against ongoing threats targeting assets within the protected system. Such centrality is reflected in the literature, where several works have been presented to propose full-fledged reaction methodologies to tackle offensive incidents’ consequences. In this direction, the work in [18] developed an immuno-based response approach based on the application of the Artificial Immune System (AIS) methodology. That is, the AIS-powered reaction is able to calculate the optimal set of atomic countermeasure to enforce on the asset within the monitored system, minimizing the risk to which those are exposed in a more than adequate time. To further contribute to this line, the paper at hand presents AISGA, a multi-objective approach that leverages the capabilities of a Genetic Algorithm (GA) to optimize the selection of the input parameters of the AIS methodology. Specifically, AISGA selects the optimal ranges of inputs that balance the tradeoff between minimizing the global risk and the execution time of the methodology. Additionally, by flooding the AIS-powered reaction with a wide range of possible inputs, AISGA intends to demonstrate the robustness of such a model. Exhaustive experiments are executed to precisely compute the optimal ranges of parameters, demonstrating that the proposed multi-objective optimization prefers a fast-but-effective reaction.

Undoubtedly, modern human digital lives are every day more and more connected, and the revolution of “everything connected” is already becoming a reality. Indeed, humans live in the age of the Internet of Things (IoT), and one of the most usual IoT contexts is a smart home. Unfortunately, such significant enhancement also means that common home devices, such as fridges, cameras, or even bulbs, are exposed to malevolent entities whose primary goal is to threaten the confidentiality, integrity, and availability of the automatically-exchanged information. Aiming at fine-tuning the protection of the smart devices, this paper proposes a novel dynamic rule management solution adaptable to the current status of the IoT environment, so to protect it against cyberattacks. Experiments demonstrated that a notable reduction in the CPU and RAM consumption was achieved when applying this novel scheme. Additionally, the number of packets processed per second increased substantially, inducing a meaningful enhancement also from a security perspective.

Abstract The evolution of integrated clinical environments (ICE) and the future generations of mobile networks brings to reality the hospitals of the future and their innovative clinical scenarios. The mobile edge computing paradigm together with network function virtualization techniques and the software-defined networking paradigm enable self-management, adaptability, and security of medical devices and data management processes making up clinical environments. However, the logical centralized approach of the SDN control plane and its protocols introduce new vulnerabilities which affect the security of the network infrastructure and the patients’ safety. The paper at hand proposes an SDN/NFV-based architecture for the mobile edge computing infrastructure to detect and mitigate cybersecurity attacks exploiting SDN vulnerabilities of ICE in real time and on-demand. A motivating example and experiments presented in this paper demonstrate the feasibility of of the proposed architecture in a realistic clinical scenario.

Due to the growth of IoT (Internet of Things) devices in different industries and markets in recent years and considering the currently insufficient protection for these devices, a security solution safeguarding IoT architectures are highly desirable. An interesting perspective for the development of security solutions is the use of an event management approach, knowing that an event may become an incident when an information asset is affected under certain circumstances. The paper at hand proposes a security solution based on the management of security events within IoT scenarios in order to accurately identify suspicious activities. To this end, different vulnerabilities found in IoT devices are described, as well as unique features that make these devices an appealing target for attacks. Finally, three IoT attack scenarios are presented, describing exploited vulnerabilities, security events generated by the attack, and accurate responses that could be launched to help decreasing the impact of the attack on IoT devices. Our analysis demonstrates that the proposed approach is suitable for protecting the IoT ecosystem, giving an adequate protection level to the IoT devices.

Summary Cloud computing is no longer the future but the present. Every day, more and more companies and service providers transfer their businesses and operations to the cloud, benefiting from its multiple advantages. Moreover, the flexibility offered by many cloud services allows to easily build sophisticated services by just composing simpler ones, rather than creating them from zero. Yet, to mitigate potential security threats and keep the maximum performance at any time, a smart selection of those composite services constitutes a key aspect. In this paper, we introduce the reader to the problem of Web service selection based on their reputation scores and subsequently present a survey on some of the most relevant reputation‐based Web service orchestration schemes for cloud computing in the literature. For each one of these approaches, a thorough analysis of their pros and cons has been performed, providing a comprehensive comparison amongst all of them leading to the conclusion that, to the best of our knowledge, there is no one single model elegantly fitting to each and every situation that could occur in such a dynamic environment like cloud computing. Finally, we present some current challenges and future research trends in the field of reputation‐based service orchestration in cloud computing. Copyright © 2013 John Wiley & Sons, Ltd.

Social network users take advantage of anonymity to share rumors or gossip about others, making it important to provide means to report offensive conduct. This article presents a proposal to automatically manage these reports. We consider not only the users’ public behavior, but also private messages between users. The automatic approach is based, in both cases, on the reporters’ reputation along with other metrics intrinsic to social networks. Promising results from adopting the proposed reporting methods on Frisber, a geolocalized social network in production, are presented as well as some experiments based on real data extracted from Frisber.

The amount of private information in the Internet is constantly increasing with the explosive growth of cloud computing and social networks. XACML is one of the most important standards for specifying access control policies for web services. The number of XACML policies grows really fast and evaluation processing time becomes longer. The XEngine approach proposes to rearrange the matching tree according to the attributes used in the target sections, but for speed reasons they only support equality of attribute values. For a fast termination the combining algorithms are transformed into a first applicable policy, which does not support obligations correctly.

SUMMARY Trust is, in some cases, being considered as a requirement in highly distributed communication scenarios. Before accessing a particular service, a trust model is then being used in these scenarios to determine if the service provider can be trusted or not. It is done usually on behalf of the final user or service customer, and with a little intervention of him or her. This is usually happening with the main aim of automatizing the process and because trust models are normally making use of reasoning mechanisms and models difficult to understand by humans. In this paper, we propose the adaptation of a bio‐inspired trust model to deal with linguistic fuzzy labels, which are closer to the human way of thinking. This Linguistic Fuzzy Trust Model also uses fuzzy reasoning. Results show that the new model keeps the accuracy of the underlying bio‐inspired trust model and the level of client satisfaction, while enhancing the interpretability of the model and thus making it closer to the final user. Copyright © 2011 John Wiley & Sons, Ltd.

Trust and reputation models research and development for distributed systems such as P2P networks, wireless sensor networks (WSNs) or multi-agent systems has arisen and taken importance in the last recent years among the international research community. However it is not always easy to check the correctness and accuracy of a model and even more, to compare it against other trust and reputation models. This paper presents TRMSim-WSN, a Java-based trust and reputation models simulator aimed to provide an easy way to test a trust and/or reputation model over WSNs and to compare it against other models. It allows the user to adjust several parameters such as the percentage of malicious nodes or the possibility of forming a collusion, among many others.