Entity linking in Arabic text presents unique challenges due to complex morphology, absence of capitalization, and dialectal variations. Traditional approaches rely on static entity representations that fail to capture contextual dynamics. This paper introduces SAGE (Semantic Adaptation of Graph Embeddings), a novel framework for Arabic entity linking that dynamically updates entity representations through a Graph Attention Network (GAT) architecture. SAGE integrates AraBERTv2 contextual embeddings with a hierarchical GAT design that processes information at both entity and type levels. The framework’s key innovation lies in its dynamic vector update mechanism that propagates semantic changes through the knowledge graph. Evaluation on Arabic news articles demonstrates that SAGE outperforms static representation methods by $28 \%$ absolute improvement in linking accuracy, with particularly significant improvements for person entities and ambiguous mentions.

International audience

International audience

Network Intrusion Detection Systems (NIDS) are mechanisms designed to improve security by monitoring networks for signs of potential intrusions. While data-driven deep learning-based NIDSs have been popular for their superior performance, they are limited by their reliance on large amounts of data, often processed in a centralized manner. Federated Learning (FL) has thus emerged as a distributed paradigm to preserve privacy and data confidentiality, reduce communication costs, and promote collaborative learning. However, FL solutions require a high degree of generalization and adaptation to data and system heterogeneity. In this paper, we introduce a new approach to enhance the generalization of deep learning models in FL-based NIDS by integrating centrality measures. These centrality measures assess the importance of nodes in a cyber-physical system, providing valuable insights into network structures. By adopting these measures within the graph constructed from source and destination devices of network flows, we aim to enhance the model's understanding of how network dynamics correlate with intrusion patterns. For our experiments, we used two public datasets: CIC-IDS-2017 and CIC-ToN-IoT. To reflect real-world network variability, we utilized a realistic federated learning setup by distributing distinct parts of the datasets among FL clients. Our approach demonstrates an improvement of over 6 % in F1-score with the use of centrality measures, surpassing the traditional baseline approach. Our findings underscore the effectiveness of integrating centrality measures in FL-based NIDS, offering enhanced intrusion detection capabilities in heterogeneous network environments.

Today, detecting outliers plays a crucial role in modern transportation systems, improving traffic management and road safety. This paper introduces a new outlier detection-based model for recognizing event patterns in Vehicular Ad hoc NETworks (VANETs). Our solution utilizes advanced techniques, combining outlier detection and multiclassification capabilities to enhance the resilience and reliability of transportation systems in dynamic and complex traffic scenarios. The approach involves three stages: data preprocessing and feature extraction, outlier detection, and multiclassification. In the first stage, an image-based dataset is transformed into a feature-based dataset after essential preprocessing operations, such as feature extraction by analyzing local patterns in the image pixels using the Local Binary Patterns (LBP) method. In the second stage, a hybrid classification model based on a neural network is proposed to identify outlier events in real-time vehicle data, followed by the employment of machine learning models to classify them as normal or abnormal. When an abnormal traffic situation is detected, the final stage utilizes multiclassification deep neural networks, specifically ResNet and Inception, to categorize events into predefined classes. Through extensive simulations using real VANET data, we have demonstrated the relevance and accuracy of our model in recognizing event patterns in traffic systems compared to other existing techniques.

The rapid economic growth along with the population concentration in urban areas have led to caused urban traffic problem. It is one of the most serious problems in big cities that people have to deal in daily life. In recent years, researchers show wide interest in overcoming traffic issues where new models and frameworks have been rapidly developed for efficient traffic management in suitable vehicular environments; the emergence of Vehicular Ad Hoc Networks (VANETs). Nevertheless, traffic flow prediction is a major challenge in VANETs that has taken much attention. Subsequently, performing accurate and real-time traffic flow prediction plays an important role in reducing traffic congestion, saving traveler time, improving traffic safety, detect accidents rapidly, and reduce infrastructure damage. In this paper, we propose an efficient traffic prediction model called prophet traffic flow predictor (Flophet) for vehicular ad hoc networks. In this model, two major enhancement on the traditional neural prophet model were done. First, we propose an efficient algorithm for predicting the traffic flow trend and, then, we implement a new future regressor component called network mobility. Through simulations in real VANET data, we show the relevance of Flophet compared to other existing models.

Big data is becoming a research focus in Intelligent Transportation Systems (ITS), which can be seen in many projects worldwide. Intelligent transportation systems will produce a large amount of data. The produced big data will profoundly impact the design and application of the ITS, which makes them safer, more efficient, and profitable. Studying big data analytics in ITS is a flourishing field. This paper aims to provide a comparative study of some recent works in Big Data analytics for Vehicular Ad Hoc Networks. The study includes reviewing some frameworks that conduct big data analytics in ITS while discussing the data source and collection methods, data analytics methods, platforms and big data analytics application categories. The comparison and implementation of five recent works, with a focus on the data collection and application layers, are followed to pick up the best approach based on the Friedman test results.

Year over year, our world becomes more and more smarter; things around us are increasingly connecting to the Internet and share information with people. This hyperconnected world leads to a new generation of Internet called the Internet of things (IoT). In such network, wireless sensor networks (WSN) constitute the eyes of the system that allow to monitor surrounding environments and sending data to a sink node for a latter analyzing. Unfortunately, the limited power energy of the sensors along with the huge amount of data collected, which are almost useless and repetitive, make saving network lifetime and decisions making are the most challenges in WSNs. In this paper, we propose two data correlation techniques followed by two scheduling mechanisms in order to switch correlated sensors into sleep/active modes. The idea behind our approach is to reduce data transmission from neighbouring sensors by eliminating data redundancy thus, saving the sensor energy and enhancing the network lifetime. The first correlation technique aims to search the similarity between datasets based on the variation between sensed data while the second one is based on the data correlation matrix. Through simulation on real sensor data, we evaluated the efficiency of our approach in terms of maintaining the node energies, ensuring zone coverage and preserving the integrity of the information.

Nowadays, the technology surrounds every corner of our lives and produces a huge amount of data about people and things' behaviours. This leads to a new sector of data analytics and decision making known as Big Data analytics era. In that era, the Internet of things (IoT) and the wireless sensor networks (WSNs) play a vital role and allow to monitor a wide number of applications, zones and environments. However, the limited resources of devices along with the redundancy among collected data makes big data collection is a major challenge for such networks. In this paper, we propose a distributed round-based prediction model dedicated to hierarchical large-scale sensor networks. First, we divide the network lifetime into a set of rounds where each round consists of several periods. At each round, each sensor collects data for some periods of the round then it sends them to the next node then, it enters into sleep mode for the other periods of the round. Upon receiving the data from each sensor, the sink uses a prediction model based on the long short-term memory (LSTM) time series in order to expect sensor data during the sleeping mode. We applied our approach on real sensor data while the obtained results show its relevance in terms of reducing data transmission and saving network lifetime.

Popularity of wireless sensor networks (WSNs) is increasing day a day where hundreds or thousands of applications are explored. In most of such applications, the need of gathering data periodically about the monitored environment beside the limited, generally irreplaceable, power sensor sources make energy conservation and big data gathering reduction two fundamental challenges in such networks. In this paper, we propose an Adaptive Distributed Data Gathering (ADiDaG) technique for saving energy in periodic WSN applications. ADiDaG works into rounds where each round consists of three phases: data gathering, sampling decision, and transmission. These phases respectively use Map reduce, longest common subsequence similarity and grouping approach in order to search data redundancy and adapt sensor sampling rate at each round. The performance of ADiDaG is evaluated based on both simulation and experimentations where the obtained results show significant energy savings and high accurate data gathering compared to existing approaches.

Data reduction is one of the most attractive way to conserve the limited energy resources of wireless sensor networks (WSNs). It aims to remove unnecessary data transmission. Therefore, data prediction and reduction mechanisms must be deployed at the source node in order to eliminate the redundant sensed data before sending them to the sink. In this paper, an energy efficient periodic distributed data reduction technique is proposed. Our technique allows each sensor node to search the variation between readings collected at each period based on the Kruskal-Wallis model. Then, the sensor selects a set of representative readings instead of sending the whole readings collected during a period to the sink. To evaluate the performance of our technique, simulations on a publicly available real sensor data followed by experiments in a real-world telosB sensor network testbed have been performed. Compared to other existing approaches, we are able to achieve up to 80% communication reduction while maintaining a high level of data accuracy.

Monitoring phenomena and environments is an emergent and required field in our today systems and applications. Hence, wireless sensor networks (WSNs) have attracted considerable attention from the research community as an efficient way to explore various kinds of environments. Sensor networks applications can be useful in different domains (terrestrial, underwater, space exploration, etc.). However, one of the major constraints in such networks is the energy consumption that increases when data transmission increases. Consequently, optimizing data transmission is one of the most significant criteria in WSNs that can conserve energy of sensors and extend network lifetime. In this article, we propose an efficient data transmission protocol that consists in two phases of data aggregation. Our proposed protocol searches, in the first phase, similarities between measures collected by each sensor. In the second phase, it uses distance-based functions to find similarity between sets of collected data. The main goal of these phases is to reduce the data transmitted from both sensors and cluster-heads (CHs) in a clustering-based scheme network. To evaluate the performance of the proposed protocol, experiments on real sensor data from both terrestrial and underwater networks have been conducted. Compared to other existing techniques, simulation and real experimentations show that our protocol can be effectively used to reduce data transmission and increase network lifetime, while still keeping data integrity of the collected data.

Data aggregation in wireless sensor networks (WSN) has been proven as an effective technique for eliminating redundancy and forwarding only the extracted information from the raw data. Furthermore, by doing so data aggregation can often reduce the communication cost and extend the whole network lifetime. In this paper we study a new prefix-suffix filtering technique for data aggregation in periodic sensor networks (PSN). We investigate the problem of finding all pair of nodes generating similar data sets. We added a new suffix frequency filter technique to the existing prefix frequency filtering. Our goal is to integrate additional filtering technique in order to decrease the latency of the aggregation phase. Our simulation results show that our technique outperforms existing prefix filtering technique in reducing energy consumption.

In-network data aggregation becomes an important technique to achieve efficient data transmission in wireless sensor networks (WSN). Energy efficiency, data latency and data accuracy are the major key elements evaluating the performance of an in-network data aggregation technique. The trade-offs among them largely depends on the specific application. For instance, prefix frequency filtering (PFF) is a good recently example for an in-network data aggregation technique that optimizing energy consumption and data accuracy. The objective of PFF is to find similar data sets generated by neighboring nodes in order to reduce redundancy of the data over the network and thus to preserve the nodes energy. Unfortunately, this technique has a heavy computational load. In this paper, we propose an enhanced new version of the PFF technique called KPFF technique. In this new technique, we propose to integrate a K-means clustering algorithm on data before applying the PFF on the generated clusters. By this way we minimize the number of comparisons to find similar data sets and thus we decrease the data latency. Experiments on real sensors data show that our new technique can significantly reduce the computational time without affecting the data aggregation performance of the PFF technique.