Attribute-based encryption provides fine-grained access control, ensuring data security in edge computing. However, existing access control schemes face issues such as high computational complexity due to bilinear pairing and key escrow with a single authority, which limit their effectiveness for lightweight terminal devices. To address these challenges, we propose a lightweight and key escrow-free access control scheme on encrypted data. This scheme replaces bilinear pairing with fast elliptic curve scalar multiplication, transferring most computational tasks to edge servers while preserving privacy, thus alleviating the computational burden on terminal devices. Additionally, we design a blockchain-based distributed key management method that allows multiple edge servers to distribute private keys collaboratively, resolving the key-escrow issue. Security analysis and performance evaluation demonstrate that our scheme effectively ensures data confidentiality, resists collusion attacks, and achieves high computational efficiency on the user side with low storage overheads, making it highly suitable for lightweight devices in edge computing.
Predicting traffic safety trends is of great significance for the realization of autonomous driving and other related areas. Current research focuses on using machine learning and deep learning models to predict traffic safety issues, such as whether a vehicle will collide or the severity of a collision. However, traffic safety is influenced by multiple factors, and relying solely on primary traffic feature data for predictions is far from sufficient. Predictions about traffic safety should consider as many factors as possible. This paper summarizes five main factors: besides primary traffic features, others include weather and environmental features, road design and configuration features, socio-economic features, and traffic violation records. Additionally, an improved meta-analysis method is introduced to quantify and evaluate the use of multiple factors, thereby identifying which factors have a greater impact on traffic safety. Meta-analysis can accurately integrate the methods and results of various studies to conclude. Thus, this paper provides a comprehensive, in-depth, and detailed review through the improved meta-analysis.
This paper represents a perspective on the design of AI chatbots with respect to the concept of method engineering. Overall, this paper includes the following sections: First, we represent primitive findings based on a set of method fragments extracted from selected papers in the domain of AI chatbot design. Then, we discuss our future work and the limitations of our research in information systems.
Grain price has a significant impact on social stable and economy development. However, the price of grain are often influenced by various complex factors, particularly the time factors. Therefore, this paper first splits the time characteristics into more fine-grained time characteristics of multiple dimensions to capture the seasonal changes, holiday effects, long-term trends, and short-term fluctuations in prices. Secondly, the isolated forest algorithm is employed to remove outliers from the dataset, ensuring the accuracy and reliability of the data set. Finally, adjust the Transformer model structure. This study removes the positional embedding layer and directly utilizes the encoding part of the Transformer model for feature extraction and representation of the input sequence, while defining the prediction layer to replace the decoder layer for numerical prediction. This design simplifies the model structure, reduces computational complexity and the number of parameters, and aligns better with the goals and requirements of regression prediction tasks. The experimental results demonstrate that the time feature decomposition method and the isolation forest algorithm significantly improve the accuracy of price prediction. Furthermore, the modified Transformer model in this study outperforms other machine learning methods and neural network algorithms in multiple evaluation metrics.
Community question answering (CQA) portals have become very popular platforms attracting numerous participants to share and acquire knowledge and information on the Internet. However, many malicious users post suggestive questions and deceptive answers to promote a target (product or service), which can extremely distort the users' decision, and make the CQA environment less credible. Although various methods have been proposed to detect the fraud content from CQA, most of them detect deceptive questions and answers separately, thus suffer the problem of interactive information missing. In this paper, we take the Question-Answer-Pair (QAP) as the detecting target and propose a GNN-based method to detect the spam activities in the CQA. Specifically, we also design a label-aware neighbor selection module based on supervised learning to select more appropriate neighbors for attributes aggregating in G NN. Extensive experiments are conducted in a real-world dataset and results demonstrate the advantage and effectiveness of the proposed model for CQA spam detection.
The prediction of taxi flow has significant research value for the optimization of overall traffic management, enhancing urban transportation efficiency, effectively reducing carbon emission, as well as improving the convenience for both passengers and drivers. It is manifest that the complexity of taxi flow data is not merely due to its inherent spatial-temporal correlation, but also impacted by various external multi-source spatial-temporal data. Despite numerous existing research efforts attempting to capture the spatial-temporal characteristics of taxi flow prediction, there are scarce results in feature integration harvested from multi-source spatial-temporal data. Hence, this research mainly explores the relationship between different multi-source spatial -temporal data and taxi flow, aiming to convert these data into external spatial-temporal correlation impact factors. Subsequently, a specially designed spatial-temporal data fusion model, Spatial-Temporal Data Fusion Network (STDFN), is employed, which categorizes these various external spatial-temporal impact factors into spatial and temporal impact factors according to their specific characteristics. Spatial and temporal influence factors are respectively processed through graph convolution and temporal convolution for feature extraction and fusion. By comparing the experimental results on real-world data with the baseline model established in this study, rigorous testing on multiple TaxiBJ datasets demonstrated an average improvement of 13.340/0 across several evaluation metrics. Furthermore, the proposed method excelled in long-term prediction tasks, showcasing higher accuracy and robustness.
In the mobile CDN environment, the information spread within the coverage of any edge server may contain sensitive information, so the sensitive information needs to be protected during the propagation process. This paper proposes to use today's more mature natural language processing (NLP) technology on the edge server. If it is detected that the information contains sensitive information through the device connection and data transmission in the coverage of an edge server, the propagation of the information will be limited according to the classification of infectious diseases. In this paper, we first assume that the sensitive information is released in the network environment, and then use SIR model to simulate the information dissemination situation. We simulate the sensitive information dissemination situation in two different situations: the edge server restricts the information dissemination and the edge server does not restrict the information dissemination, and compare the experimental results of the two different situations. In this paper, dolphin network data set and simulation data set are used for experimental verification. The experimental results show that, compared with the traditional cloud computing framework, this method can effectively inhibit the spread of sensitive information in the mobile CDN network framework and limit the spread of sensitive information.
The selection of the optimal join order is critical for the efficiency of join query execution. The addition of new tables or columns necessitates the retraining of existing reinforcement learning models from scratch, consuming a significant amount of time and computational resources. To address this problem, we proposes a transfer reinforcement learning method that utilizes policy distillation techniques to optimize join order selection. The proposed algorithm allows the policy from a trained model (teacher model) to be transferred to a new model (student model), significantly reducing training costs associated with model changes. Additionally, we introduced a matrix representing the join sequence to address the issue of different join orders in two different join trees resulting in identical encoding. The experiments demonstrate that this method reduces training time and improves the speed of query execution compared to existing algorithms.
Roadside Units (RSUs) constitute a vital component of Vehicular Ad Hoc Network (VANET) due to their primary role in gathering vehicle information. However, as indicated in the literature, they are vulnerable to Denial of Service (DoS) attacks. Existing research has proposed several methods for detecting DoS attacks in RSUs, while subsequent steps such as mitigation and elimination remain unexplored. In this work, we propose a novel DoS Attack defense method for roadside units. We employ re-authentication to identify Sybil vehicles and discard the corresponding packets, thereby mitigating the attacks. Subsequently, we calculate the location of the actual malicious vehicle using a triangulation algorithm, in cooperation with traffic cameras and a cloud platform. This allows us to confirm the attacker and eliminate the DoS attacks. Simulation results demonstrate that our proposed method effectively defends against DoS attacks and successfully traces the attackers. Our method is a supplement to existing detection techniques.
Critical infrastructures increasingly rely on unmanned aerial vehicles (UAVs) for inspection tasks. The significance of different components within these infrastructures is subject to variability and can be influenced by external factors. The principal aim of routine UAV inspections is to ensure differentiated coverage of pivotal sections with minimal energy consumption. However, extant research on UAV path coverage fails to fully account for the variability and dynamic shifts in regional significance and their impact on coverage efficacy. This paper presents an energy-aware collaborative coverage policy for UAVs, designated as EA-MATD3.EA-MATD3 employs a dynamic weight region partitioning method tailored to real-world environments and addresses the action selection challenge for UAVs using a discrete Partially Observable Markov Decision Process (Dec-POMDP). By amalgamating MATD3 with stacked LSTM, this approach mitigates redundant path overlaps and unnecessary action replication across multiple agents, thus optimizing coverage and diminishing energy usage. Simulation outcomes demonstrate that EA-MATD3 reduces energy consumption by an average of 9.65% relative to the Greedy, MADDPG, and MATD3 algorithms while sustaining a superior coverage rate.
Over the years, the Vehicle Routing Problem (VRP) has been extensively studied. As real-world scenarios and technol-ogy evolve, researchers are increasingly integrating drones with trucks, leading to the development of the Two-Echelon Vehicle Routing Problem (2EVRP). The 2EVRP involves dividing the route plan into two segments: the first echelon operated by trucks and the second by drones. With the introduction of lightweight, cost-effective delivery solutions like drones, the 2EVRP faces challenges such as limited drone battery life and adherence to public safety policies. This survey examines the literature on traditional exact and heuristic algorithms alongside contem-porary machine learning approaches, including supervised and reinforcement learning methods. Additionally, we review current real-world applications of drone delivery. Finally, we highlight several future research directions for 2EVRP with truck and drone collaborations.
The escalating frequency and sophistication of cyberattacks underscore the urgent need for robust threat intelligence. This paper proposes a novel approach to harnessing the wealth of information on Twitter for timely cyber threat detection. By leveraging natural language processing and Deep learning, specifically Iterated Dilated Convolutional Neural Networks (IDCNN) and Bidirectional Long Short-Term Memory (BiLSTM), we developed a IDCNN-BiLSTM learning model capable of accurately identifying cyber threats from Twitter data. Our model was trained on a comprehensive dataset of threat-related tweets and demonstrated superior performance compared to existing methods. This research contributes to the development of advanced cyber threat intelligence systems by providing a scalable and effective solution for real-time threat detection.
Sequential recommendation aims to capture the dynamic changes in user interests by leveraging their historical behavior sequences for making recommendations. Existing methods often rely on explicit item IDs or generic textual features, but they struggle with cold start scenarios and adapting to new datasets. In this paper, we propose a novel approach called Text-based Multi-pair Contrastive Learning Bidirectional Transformer for Sequential Recommendation (TMCBiT). This method takes sequences of users' historical interactions in the form of key-value text pairs as input, enhancing representation capability through multi-layer embeddings. It employs a bidirectional Transformer model with a long-range attention mechanism, integrating random masking training with multi-pair contrastive learning for joint optimization. Experimental results demonstrate that our proposed method effectively alleviates issues related to cold starts and new dataset adaptation, significantly improving recommendation performance. This study provides a viable path for enhancing model performance in recommendation systems using deep learning techniques, offering valuable guidance for developing more intelligent, accurate, and personalized recommendation systems.
Identifying elephant flows in a network is crucial for network traffic measurement, significantly impacting congestion control optimization, anomaly detection, and traffic engineering. With increasing network link rates and limited on-chip storage space, existing algorithms face severe challenges in maintaining measurement accuracy. To address these issues, Sketch data structures have emerged, allowing the recording of all traffic characteristics within limited on-chip storage. However, due to the approximate nature of Sketch, its measurement accuracy often falls short of high precision requirements. In this paper, we propose a Sketch algorithm combined with an SDN (Software-defined Networking) controller, termed CTS (Combine the Two-Stage SDN) Sketch, to enhance the accuracy of identifying top-k elephant flows through additional filtering. Furthermore, we conduct measurements exclusively at edge switches, uploading the distributed Sketch measurement results to the controller to reduce the load on central devices. By leveraging the advantages of edge computing, we capture and process traffic information promptly. Experimental results show that CTS Sketch achieves an accuracy rate of 99.99% with relatively small memory capacity, reducing error by approximately three orders of magnitude compared to existing algorithms.
Cybersickness induced by virtual reality (VR) ap-plications remains one of the main obstacles to its development. Despite extensive research on reducing cybersickness, there is a lack of non-invasive methods to predict the severity of users' cybersickness in advance. Considering the advancements in eye-tracking technology within VR head-mounted displays and previous studies on the correlation between blinking behavior and cybersickness, this study aims to propose a method for predicting users' future blinking behavior, thereby providing a foundation for subsequent non-invasive cybersickness prediction by leveraging the correlation between blinking and cybersickness. Based on the encoder-decoder architecture, this study compre-hensively considers various ocular movement features of users during a VR experience and develops a blink prediction model using a dataset that records eye movement information from 23 participants during a virtual driving experience. The results show that the model can effectively predict the number of blinks within a future one-second interval. Furthermore, since the number of blinks at the same level of cybersickness can vary rather than remain fixed, the model also predicts the range of blink counts over the next 10 seconds, which can provide a basis for future work on predicting cybersickness.
In contemporary education systems, educational data is a core element in driving educational advancement and optimizing management. Most educational data is stored in the cloud as videos, images, or text, which facilitates its storage and dissemination. However, there are two significant challenges: First, existing data compliance detection systems may be vulnerable to attacks where malicious actors inject unqualified data into detection models, resulting in the upload of non-compliant educational data. Second, there is the issue of incomplete data storage. Data in cloud computing environments might be lost or corrupted due to various issues, posing a threat to the utilization of educational data. To address these challenges, this paper proposes a blockchain-based auditing scheme for educational data supporting trusted detection scheme(BAS-EDTD). This scheme uses smart contracts to randomly select AI detection modules for data inspection, ensuring that uploaded data is harmless and generating a detection report stored in the cloud. Additionally, a Trusted Third Party (TPA) assesses the reputation of AI detection modules to prevent those in untrusted cloud environments from injecting non-compliant data and affecting detection efficiency. Simultaneously, we adopt certificateless aggregate signature technology to perform data integrity auditing, reduce storage and computing overheads, support batch verification, and improve verification efficiency. Security proofs indicate that this scheme can effectively resist forgery attacks and demonstrates lower computing and communication overheads compared to existing schemes in performance analyses.
Contemporary research on sequential recommender systems primarily focuses on optimizing accuracy, but recommendation diversity is often overlooked. This concept involves recommending diverse items to users, providing them with broader choices and preventing content from being too similar. There is typically a trade-off between accuracy and diversity, as increasing diversity can reduce accuracy by recommending items that are less similar to a user's previous preferences. While trust, as auxiliary information, is often used to increase recommendation accuracy, the potential of leveraging trust to enhance the diversity of sequential recommendations remains largely unexplored. In this paper, we propose a trust-aware recommendation method aimed at enhancing diversity in sequential recommendations while maintaining accuracy. In this paper, we propose a trust-aware recommendation method aimed at enhancing diversity in sequential recommendations while minimizing any impact on accuracy. Experiments on two public datasets from various domains demonstrate that the proposed method outperforms baseline methods in both overall and individual diversity while maintaining a comparable level of accuracy.
The Industrial Internet of Things (IIoT) represents a promising application of the Internet of Things (IoT) in industry, offering significant potential for enhancing manufacturing efficiency and enabling smart production. In the context of the IIoT, data is stored in the form of ciphertext on cloud servers. This presents a challenge in terms of locating the data within the ciphertext. Public key searchable encryption (PEKS) represents a cryptographic primitive that may be employed to address this challenge. However, the traditional PEKS scheme is susceptible to a security threat in the form of a keyword guessing attack (KGA), in addition to presenting challenges associated with certificate management and key escrow. In this paper, we introduce an innovative Certificateless Searchable Encryption (CLSE) scheme that combines the advantages of certificate-less encryption techniques, effectively protects against Key Generation Attack (KGA), and solves the key escrow problem. A detailed security analysis and performance evaluation confirms the suitability of the solution for real-world Industrial Internet of Things (IIoT) applications. It properly handles the above security challenges while ensuring efficient computation and communication.
In recent years, federated learning (FL), a privacy-aware distributed learning paradigm, has been facilitating numerous cloud and mobile edge systems. Although FL proves effective in enhancing edge services, the complex circumstances of edge computing (e.g., data heterogeneity, limited communication capacity, and arbitrary service unavailability) can adversely affect FL applications, resulting in slow convergence and low-quality services. Existing FL methods merely address some of these issues, rendering them unsuitable for edge applications. To address these challenges, we propose a two-fold method, FedTSB, to alleviate both sustained bias and temporary bias, enabling rapid and stable training in complex edge computing scenarios. Experimental results on two datasets under three heterogeneous scenarios demonstrate that our proposed FedTSB outperforms four baseline methods in terms of both performance and efficiency metrics.
