With the propagation of cyberbullying in social networks as a trending subject, cyberbullying detection has become a social problem that researchers are concerned about. Developing intelligent models and systems helps detect cyberbullying automatically. This work focuses on text-based cyberbullying detection because it is the commonly used information carrier in social networks and is the widely used feature in this regard studies. Motivated by the documented success of neural networks, we propose a complete model combining the bidirectional gated recurrent unit (Bi-GRU) and the self-attention mechanism. In detail, we introduce the design of a GRU cell and Bi-GRU’s advantage for learning the underlying relationships between words from both directions. Besides, we present the design of the self-attention mechanism and the benefit of this joining for achieving a greater performance of cyberbullying classification tasks. The proposed model could address the limitation of the vanishing and exploding gradient problems. We avoid using oversampling or downsampling on experimental data which could result in the overestimation of evaluation. We conduct a comparative assessment on two commonly used datasets, and the results show that our proposed method outperformed baselines in all evaluation metrics.

Bluetooth Low-Energy (BLE) beacons-based indoor positioning is a promising method for indoor positioning, especially in applications of position-based services (PbS). It has low deployment cost and it is suitable for a wide range of mobile devices. Existing BLE beacon-based positioning methods can be categorized as range-based methods and fingerprinting-based methods. For range-based methods, the positions of the beacons should be known before positioning. For fingerprinting-based methods, a pre-requisite is the reference fingerprinting map (RFM). Many existing methods focus on how to perform the positioning assuming the beacon positions or RFM are known. However, in practical applications, determining the beacon positions or RFM in the indoor environment is normally a difficult task. This paper proposed an efficient and graph optimization-based way for estimating the beacon positions and the RFM, which combines the range-based method and the fingerprinting-based method. The method exists without need for any dedicated surveying instruments. A user equipped with a BLE-enabled mobile device walks in the region collecting inertial readings and BLE received signal strength indication (RSSI) readings. The inertial measurements are processed through the pedestrian dead reckoning (PDR) method to generate the constraints at adjacent poses. In addition, the BLE fingerprints are adopted to generate constraints between poses (with similar fingerprints) and the RSSIs are adopted to generate distance constraints between the poses and the beacon positions (according to a pre-defined path-loss model). The constraints are then adopted to form a cost function with a least square structure. By minimizing the cost function, the optimal user poses at different times and the beacon positions are estimated. In addition, the RFM can be generated through the pose estimations. Experiments are carried out, which validates that the proposed method for estimating the pre-requisites (including beacon positions and the RFM). These estimated pre-requisites are of sufficient quality for both range-based and fingerprinting-based positioning.

New aerial sensors and platforms (e.g., unmanned aerial vehicles (UAVs)) are capable of providing ultra-high resolution remote sensing data (less than a 30-cm ground sampling distance (GSD)). This type of data is an important source for interpreting sub-building level objects; however, it has not yet been explored. The large-scale differences of urban objects, the high spectral variability and the large perspective effect bring difficulties to the design of descriptive features. Therefore, features representing the spatial information of the objects are essential for dealing with the spectral ambiguity. In this paper, we proposed a dual morphology top-hat profile (DMTHP) using both morphology reconstruction and erosion with different granularities. Due to the high dimensional feature space, we have proposed an adaptive scale selection procedure to reduce the feature dimension according to the training samples. The DMTHP is extracted from both images and Digital Surface Models (DSM) to obtain complimentary information. The random forest classifier is used to classify the features hierarchically. Quantitative experimental results on aerial images with 9-cm and UAV images with 5-cm GSD are performed. Under our experiments, improvements of 10% and 2% in overall accuracy are obtained in comparison with the well-known differential morphological profile (DMP) feature, and superior performance is observed over other tested features. Large format data with 20,000 × 20,000 pixels are used to perform a qualitative experiment using the proposed method, which shows its promising potential. The experiments also demonstrate that the DSM information has greatly enhanced the classification accuracy. In the best case in our experiment, it gives rise to a classification accuracy from 63.93% (spectral information only) to 94.48% (the proposed method).

With the development of artificial intelligence, machine learning algorithms and deep learning algorithms are widely applied to attack detection models. Adversarial attacks against artificial intelligence models become inevitable problems when there is a lack of research on the cross-site scripting (XSS) attack detection model for defense against attacks. It is extremely important to design a method that can effectively improve the detection model against attack. In this paper, we present a method based on reinforcement learning (called RLXSS), which aims to optimize the XSS detection model to defend against adversarial attacks. First, the adversarial samples of the detection model are mined by the adversarial attack model based on reinforcement learning. Secondly, the detection model and the adversarial model are alternately trained. After each round, the newly-excavated adversarial samples are marked as a malicious sample and are used to retrain the detection model. Experimental results show that the proposed RLXSS model can successfully mine adversarial samples that escape black-box and white-box detection and retain aggressive features. What is more, by alternately training the detection model and the confrontation attack model, the escape rate of the detection model is continuously reduced, which indicates that the model can improve the ability of the detection model to defend against attacks.

The explosive growth of cyber attacks today, such as malware, spam, and intrusions, has caused severe consequences on society. Securing cyberspace has become a great concern for organizations and governments. Traditional machine learning based methods are extensively used in detecting cyber threats, but they hardly model the correlations between real-world cyber entities. In recent years, with the proliferation of graph mining techniques, many researchers have investigated these techniques for capturing correlations between cyber entities and achieving high performance. It is imperative to summarize existing graph-based cybersecurity solutions to provide a guide for future studies. Therefore, as a key contribution of this work, we provide a comprehensive review of graph mining for cybersecurity, including an overview of cybersecurity tasks, the typical graph mining techniques, and the general process of applying them to cybersecurity, as well as various solutions for different cybersecurity tasks. For each task, we probe into relevant methods and highlight the graph types, graph approaches, and task levels in their modeling. Furthermore, we collect open datasets and toolkits for graph-based cybersecurity. Finally, we present an outlook on the potential directions of this field for future research.

In real-world scenarios, the number of phishing and benign emails is usually imbalanced, leading to traditional machine learning or deep learning algorithms being biased towards benign emails and misclassifying phishing emails. Few studies take measures to address the imbalance between them, which significantly threatens people’s financial and information security. To mitigate the impact of imbalance on the model and enhance the detection performance of phishing emails, this paper proposes two new algorithms with undersampling: the Fisher–Markov-based phishing ensemble detection (FMPED) method and the Fisher–Markov–Markov-based phishing ensemble detection (FMMPED) method. The algorithms first remove benign emails in overlapping areas, then undersample the remaining benign emails, and finally, combine the retained benign emails with phishing emails into a new training set, using ensemble learning algorithms for training and classification. Experimental results have demonstrated that the proposed algorithms outperform other machine learning and deep learning algorithms, achieving an F1-score of 0.9945, an accuracy of 0.9945, an AUC of 0.9828, and a G-mean of 0.9827.