With continuous growth of IoT applications, service failures are quite inevitable. Due to the complexity and dynamics of IoT services, the root cause analysis (RCA) following an alert can assist in quickly resolving the possible faults. However, the time scales of metrics (e.g., CPU utilization, memory usage) generated by microservices and the dynamic topologies generated by calls between the Application Program Interfaces (APIs) are different. Moreover, the status of devices is an important aspect of RCA in IoT. All these make it extremely challenging to learn failure features of microservice metrics and API calls. Therefore, we propose a novel framework for collaborative identification of root cause analysis (CIRCA) to identify the most potential root cause path with the highest fault scores (weights). In detail, we use both microservice-level and API-level root cause identification (RCI) models to obtain the node fault score in the path. Since we prove the root cause path inference problem is an NP-hard problem, and we then propose a topology-based weighted variable neighborhood search (TWVNS) algorithm and infer the optimal root cause path from two-level scores and call topologies. Our experiments demonstrate CIRCA achieves satisfactory results of RCI and path inference on four public datasets.

Multivariate time series forecasting has a wide range of applications in the fields of computer supported cooperative work. In recent years, deep learning models like transformers, have been widely used in time series forecasting. However, traditional transformers mainly capture short-term dependencies (point to point) within single-variable sequences and struggle with capturing cross-variable and multi-scale dependencies for long-term forecasting. To address this issue, we propose a novel transformer model called MLPT, an MLP-Enhanced Patch Transformer for Multivariate Time Series Forecasting. Our model extracts global features, reconnects them with patches and then calculates both inter and intra attenions, being capable of capturing dependencies in time series from both inter and intra variables, and models these dependencies globally, which makes our model outperform in long-term forecasting. We evaluate our method using seven real-world public datasets, and the experimental results demonstrate that our approach has improved over existing methods in the vast majority of scenarios, with an overall improvement of above 5% on the accuracy. At the same time, the time cost is reduced to one-fourth compared with MTGNN.

With the rapid advancement of information technology, the volume of log data generated by information systems has expanded significantly. The effective detection and processing of anomalies in log sequences has emerged as a crucial research topic. Recent studies have successfully applied pre-trained models for anomaly detection in log sequences. However, due to the semi-structured nature of logs and the presence of domain-specific vocabulary, along with the diverse logging styles of different vendors, those models struggled to provide a comprehensive understanding and unified representation of logs. Additionally, current research primarily focused on log template sequences, neglecting the impact of variable values and contextual relationships among logs. In this paper, we propose LlmBertLog, a BERT-based log sequence anomaly detection model guided by large language models. We first utilize a large language model to transform a limited set of raw logs into natural language descriptions as sample data. Subsequently, we induct three novel pre-training tasks designed to enhance BERT's comprehension and representation capabilities through methods such as semantic augmentation and semantic masking. We then integrate a Transformer with positional encoding and fine-tune LlmBertLog for log sequence anomily detection task, thereby improving its performance in log sequence content mining. Extensive experiments demonstrate that LlmBertLog achieves F1 scores of 0.994 on the BGL dataset and 0.998 on the HDFS dataset for log sequence anomaly detection, surpassing the current best benchmarks by 2.3% and 5.5%, respectively.

Multivariate time series forecasting has a wide range of applications in scenarios such as wireless communication and mobile computing. Recent research has demonstrated that transformer architectures are highly effective for time series forecasting, with smaller models achieving impressive results across multiple public datasets. However, smaller models tend to underperform in scenarios where data is scarce, such as in zero-shot prediction tasks. In recent years, with the rise of large language models, there has been growing interest in developing large models for time series prediction, leading to a series of advancements in this area. However, the current models often suffer from issues such as neglecting the inter-dependencies between multi-dimensional variables and their operations are too complicated. In this paper, focusing on zero-shot prediction, we enhance the Timer model by efficiently and adaptively capturing multivariate dependencies and frequency domain dynamics. Experiments show that the zero-shot prediction capability of our model has significantly improved on multiple public datasets, with an decrease in MSE for over 10%.

Anomaly detection of sensor data is crucial to ensure the stability and effectiveness of Internet of Things (IoT) system. The task requires high accuracy and low latency, which makes distributed anomaly detection gradually become a research hotspot. Edge AI networks further enhance the computing power of edge servers, allowing distributed anomaly detection to be gradually applied to complex scenarios under the Industrial IoT (IIoT), such as smart factories. However, in such cases, the data reported by different sensors in different fields are not independent and identically distributed (non-IID). Simultaneously, during unsupervised training, anomaly data mixed into the training data will reduce the recognition ability of the anomaly detector. To address these challenges, we propose an unsupervised distributed anomaly detection framework. On the cloud, we train an unsupervised anomaly detection model with global factors, using global and local factors to learn the distribution patterns of different fields. At the edge, a multidimensional threshold and its automatic selection algorithm are proposed to overcome the problem of decreased anomaly recognition ability introduced by anomaly training data. Extensive experiments on six datasets show that our approach outperforms SOTA methods in F1-score, can detect anomalies with high accuracy and efficiency in distributed IoT scenarios.

Encrypted traffic classification is a challenging task characterized by its invisibility of content, class imbalance, and limited number of labeled samples. In current research, the application of pre-trained language models has achieved significant success. However, employing language models for encrypted traffic classification on resource-constrained devices requires reducing model size and inference cost while preserving accuracy. In this paper, we proposed an unsupervised, task-agnostic, transformer-based language model knowledge distillation method for encrypted traffic data, called NetKD. It leverages Multi-Head Self-Attention Relation Alignment, Masked BURST Model, and Same-origin BURST Prediction tasks, to effectively transfer the byte-level contextual knowledge about encrypted traffic data learned by a large-scale "teacher" BERT model, to a compact "student" NetKD-BERT model. The student model only requires a small amount of supervised data for fine-tuning then can be applied to downstream tasks. Experiments show that compared to the state-of-the-art baseline model, our model uses only 4.61% of the parameters, achieves on average 99.10% of its F1 score in 4 downstream tasks, with a classification speed of 0.93ms/packet. Its memory footprint is as low as 7.33% of the baseline model’s, leading to a significant 14.3 times faster inference speed.

Nowadays, due to the increasingly complex and huge data flow, anomaly detection has become an important technology to ensure the high security and reliability of the systems. It helps identify unusual observations in the data, thereby protecting safety and reducing risk. However, in the field of multivariate temporal anomaly detection, there are still some difficulties in spatio-temporal correlation mining of variables, generalization ability in dealing with complex temporal sequences, and feature learning under different granularity. Therefore, we propose an unsupervised anomaly detection method for multivariate time series based on contrastive learning and Autoformer autoencoder. In order to fully explore the spatio-temporal correlation of time series data, this paper uses Graph Attention Network and Auto-Correlation to jointly learn spatio-temporal correlation, and obtain positive and negative pairs through time-domain and frequency-domain data augmentation. Finally, the Autoformer autoencoder is used for anomaly detection in a predictive manner. In the loss function section, a joint loss function is formed by introducing contrastive learning losses at the time step and time window granularity, combined with prediction loss, to improve the learning effect. This paper investigates the performance of the proposed method through experiments on the SMAP and MSL public datasets, and achieves better results than baseline methods.

Chinese text Classification models are vulnerable to adversarial attacks. Based on Chinese language features such as phonology and glyphs, attackers can modify serval words or characters and perturb the results of the classification model without affecting the semantics of the sentence. Adversarial examples are becoming serious challenges to the robustness of the classification model, even for the state of the art models such as PLMs. Therefore, it is necessary to effectively improve the robustness of text classification models with Chinese language features. In this paper, we propose an efficient defense method CWordDefender to address the adversarial robustness problem in Chinese text classification tasks. We extract multimodal information based on Chinese features and fine-tune PLMs with contrastive-adversarial learning. Experimental results show that CWordDefender is superior to the baseline model by at least 5% in accuracy and has lower infer time.

With the fast development of Internet of Things (IoT), anomaly detection has recently become a common concern for IoT smart applications. The circuit detection mode of all services to detect application anomaly is widely adopted. However, regularly collecting key performance indicators (KPIs) of all services under different clouds is a time-consuming task. Moreover, too many alerts in a short period of time will affect the processing speed of engineers. Recently, some researches of inferring the key service on the call paths can solve the above problems. However, it is still a challenge to locate dynamic and scattered key services accurately. In this article, we propose a service inference-based anomaly detection approach (SIADA), which integrates three sources of logs: 1) call; 2) business; and 3) metric. SIADA leverages the deep graph representation with context-aware multigraph fusion based on a recurrent encoder. This should infer key services and adopt variational autoencoder (VAE) with the flow model to detect multivariate time-series anomalies for key services. We have conducted extensive experiments on the public data set MicroSS. SIADA achieves the best average accuracy of 92% in service inference and the best average F1-score of 0.98 in anomaly detection, which has improved by 12.17% and 6.42% compared with the best SOTA baseline, respectively. Moreover, the total detection time, network transmission, and average alert number are reduced by 42.12%, 81.87%, and 83.5%, respectively.

With the rapid development of cloud-network integration, in order to ensure the normal operation of devices, networks, and services in the cloud, anomaly detection algorithms have received increasing attention. The existing anomaly detection methods for multivariate time series, on the one hand, tend to identify situations where multiple metrics deviate significantly from the normal patterns as anomalies, while ignoring anomalies with small deviations from a small number of metrics but very critical to the system, which has a certain tendency. On the other hand, when selecting the window length to extract time series features, it is difficult to take into account both the local and the global data when extracting the data features in the sliding window. Therefore, this paper proposes an anomaly detection method for multivariate time series based on the cross window and probability balance by combining VAE and AE to weaken the tendency of the existing classical algorithm, and can extract the global information and detailed information of the data in the window at the same time. According to the experimental results, the method in this paper has better performance than the best recently baseline, and the F1-score is improved by about 6%.

Open Information Extraction (Open IE) provides a method to extract triplets from text, which has become a research frontier and hot topic in recent years. However, the traditional open information extraction approaches have low precision and significant costs, which heavily rely on the artificially defined extraction paradigm. In this paper, we propose ZoIE, a zero-shot open information extraction model. First, we set positive and negative samples and use the pretraining method based on contrastive learning to train a language model. Then, we use the self-attention weight matrix to perform triplet extraction by using the method of nucleus sampling beam search. Finally, we construct a triplet fine-sorting strategy based on the loss function and select the optimal triplet from the candidate triplets. Experimental results on serveral datasets show that our proposed approach achieves better performance than other baselines.

With the promotion of cloud computing, a large number of hardware and software systems in the cloud bring massive and complex operation and maintenance (O&M) work. To ensure the O&M efficiency of IT infrastructures, it is necessary to implement automatic and reliable scheduling for the directed acyclic graph (DAG) workflow which is composed of multiple O&M tasks. Considering the changing status of networks and machines in the cloud and the position constraints that some tasks must be executed on the specified machines in some O&M scenarios, we propose a novel workflow scheduling approach based on Deep Reinforcement Learning (DRL) to minimize the workflow execution makespan and implement the fault tolerance with the position constraints of tasks execution. In our proposal, we first design a fault-tolerant mechanism according to the reliability requirement and the probability distributions of the machine failure parameters with consideration of different failure rates in the heterogeneous environment. Then, we employ proximal policy optimization (PPO) to optimize the task scheduling strategy and ensure the strategy to satisfy the position constraints of tasks execution by action masking in proximal policy optimization. The experimental results show that our proposal can effectively reduce the makespan of the fault-tolerant workflow on the premise of 99.9% reliability.

Joint extraction of entity and relation is a basic task in the field of natural language processing. Existing methods have achieved good result, but there are still some limitations, such as span-based extraction cannot solve overlapping problems well, and redundant relation calculation leads to many invalid operations. To solve these problems, we propose a novel RelationSpecific Triple Tagging and Scoring Model (RS-TTS) for the joint extraction of entity and relation. Specifically, the model is composed of three parts: we use a relation judgment module to predict all potential relations to prevent computational redundancy; then a boundary smoothing mechanism is introduced to the entity pair extraction, which reallocates the probability of the ground truth entity to its surrounding tokens, thus preventing the model from being overconfident; finally, an efficient tagging and scoring strategy is used to decode entity. Extensive experiments show that our model performs better than the state-of-the-art baseline on the public benchmark dataset. F1-scores on the four datasets are improved, especially on WebNLG and WebNLG∗, which are improved by 1.7 and 1.1 respectively.

With the rapid development of enterprise digital transformation and the widespread use of cloud-native architecture applications, the complexity of enterprise-level systems is getting higher and higher. This requires the engineers to be able to locate and repair alert incidents accurately and quickly to ensure the stability of complex network equipment. In this paper, we propose a novel alert prediction framework, ZAlert, which can predict the occurrence of future alerts and locate the equipment that may cause alerts based on real-time alarm data. First, ZAlert extracts text and statistical features from alarm data to build a high-performance comprehensive learning model to predict the category of alert that may occur in the future with an average performance of 0.81 on F1-score. Then, we use the knowledge graph based on the CMDB(Configuration Management Database) to search for alarm devices. In this way, engineers can quickly find devices that may cause alarms, and improve the the overall efficiency of engineers’ operation and maintenance.

Multiaccess edge computing has emerged as a powerful paradigm for increasing the computation performance of mobile devices (MDs). Applying nonorthogonal multiple access (NOMA) to MEC can further improve the spectrum efficiency and reduce offloading delays caused by the upload congestion. In this article, we examine the joint computation offloading and resource allocation problem in the NOMA–MEC system, which benefits from the combination of NOMA and MEC. Our optimization objective is to minimize the computational overhead (the weighted sum of the execution delay and the energy consumption) in dynamic environments with time-varying wireless fading channels. The optimization problem is formulated as a mixed-integer programming (MIP), which involves jointly optimizing the task offloading decisions, channel assignment, and transmit power allocation. To solve such an optimization problem, we formalize the task offloading and the resource allocation as a Markov decision process (MDP). Then, we propose a deep reinforcement learning (DRL)-based approach, which combines multiple deep neural networks (DNNs) to directly approximate different statistical models for continuous and discrete control. The simulation results demonstrate that the proposed approach can rapidly converge and efficiently decrease the total computational overhead compared to other baseline approaches in different scenarios.

One of the most challenging problems in the question answering (QA) area is the lack of high-quality labeled data. However, the cost of manually labeling a question-answer (QA) pair from the target text is very high. One way to solve this problem is to automatically generate QA pairs from the target text. In this paper, we propose the Batch Normalization conditional VAE-QA pair generation (BNCVAE-QAG) model to generate QA pairs for a given text. First, we employ the Batch Normalization (BN) layer to prevent Kullback-Leibler (KL) divergence from disappearing. We also design modules to extract spatiotemporal features from text contents, in addition, the self-attention mechanism is employed in the question decoder, which improves the accuracy and recall of results. Furthermore, we propose a question generalization mechanism to generate more QA pairs. We evaluate our BNCVAE-QAG model on several datasets. The experimental results show that our model has achieved an impressive performance improvement than the baseline.

With the rapid increase of users, online meeting platforms have accumulated massive meeting transcripts. However, it is still a challenge for users to quickly master the chief information and manage the meetings, despite there are already some useful text summarization models. In this paper, a Knowledge Graph-based Meeting Summarization Framework is proposed to tackle this challenge. First, a two-layers meeting domain Knowledge Graph is developed to integrate more information of meetings. Based on which, an encoder-decoder architecture is utilized to summarize meetings. For encoding meetings, a structural-level and semantic-level embedding strategy is considered, concretely, the Knowledge Graph is embedded to obtain the structural information, an interaction intention recognition model and a two-level transformer mechanism are devised to get the semantic information. Finally, the structural information and semantic information are combined and fed into the decoding network to generate meeting summaries. Extensive experiments on the Chinese meeting dataset show that our summarization framework outperforms other state-of-the-art models.

As the school’s surveillance cameras generate a large amount of video data every day, analyzing these video images to promote the development of intelligent campuses has become a research hot spot. By classifying the body postures, we can analyze the attention of students during the learning process. Many researchers have proposed methods to identify the posture of the human body in images. However, due to the occlusion and angle of the video data, these methods are not suitable for classroom images. In this paper, we propose S2FCM, a two-stage fine classification model to analyze the attention of students. We first propose a two-branch network to classify body posture. To deal with the occlusion in the image, we use the skeleton heat map as the input of the graph convolutional structure. In the second stage, we propose a position-robust network to classify the head posture. To solve the problems caused by the position difference, we design a position correction module. Experiments show that the macro-f1-score of our classification model is at least 20% higher than that of the other two general methods.

Mobile edge computing (MEC) has been regarded as a promising paradigm for increasing the computing capacity of mobile devices (MDs) by offloading tasks to edge servers. Non-orthogonal multiple access (NOMA) is a critical multiple access technique that allows many MDs to transmit on the same resource block simultaneously. Attracted by the enormous benefits of combining NOMA and MEC, we investigate the dynamic computation offloading in a multi-device multi-server NOMA-MEC system. We consider the partial offloading policy such that MDs can offload a portion of the task to the edge server for execution. To minimize the overall computation delay and energy consumption, we formulate a mixed integer programming (MIP) problem to jointly optimize edge server selection and offloading task ratio. Solving the optimization problem with a discrete-continuous hybrid action space is not straightforward since most existing deep reinforcement learning (DRL) algorithms are only applicable to discrete or continuous action spaces. We present the hybrid advantage actor-critic (HA2C) approach, which employs an actor-critic architecture consisting of two par-allel actor networks and a critic network, to tackle this problem. Specifically, the discrete actor and continuous actor networks based on deep neural networks (DNNs) determine MEC server selection and offloading ratio, respectively. The critic network evaluates the current state value, and the advantage function is computed for the discrete and continuous network parameter updates. Experimental results show that the proposed algorithm is superior to conventional DRL algorithms that convert the hybrid action space into a unified homogeneous action space.

With the development of visual programming languages, researchers pay attention to the automatic evaluation of visual projects. Previous work focus on the code evaluation but ignored another essential part—the visualization results. Scratch is a widely used programming platform, and projects created on it are displayed in the form of cartoon clips. It is valuable to explore the visual aesthetics embodied in these clips to fill the gap in the assessment system. We propose a model that predicts the human view scores of cartoon clips created on Scratch. Our method is divided into two steps to evaluate the aesthetic of the sequence images that compose cartoon clips. First, we train an image classification network to predict the relative aesthetics of individual images. Then we construct an aesthetic space for the sequence image and improve the rating within a specific range. We put forward ScratchGAN to generate a Scratch-cartoon-style aesthetic analysis data set for training the classification network. Experimental results show that our Generative Adversarial Network framework can well transform photos into a Scratch-cartoon style. The single image assessment network can generate predictions that fit human cartoon aesthetic opinions. Our method achieves satisfactory results in the aesthetic evaluation of sequence cartoon images.

Nowadays, Scratch, as a widely-used educational programming platform, has gathered a huge number of programming users all over the world. Facing massive programming resources, how to make satisfactory programming recommendations has attracted increasing attention, especially on explainable recommendations. Existing Scratch recommendation systems overlook to provide why a project is recommended, which prevents users from making better decisions and trusting in the system. To resolve this problem, we design the Scratch-RL, an explainable reinforcement learning framework over knowledge graphs for Scratch recommendation. First, we devise a preference-driven Actor-Critic network to simulate users' local preferences and explore the potential interested projects along the reasoning paths. In the Actor-Critic network, we elaborate a preference state function, a preference-based reward function, and a preference-conditional action pruning strategy for the agent. Then, we leverage a directive discriminator network to help evaluate the correctness of recommendations from the agent and return an extra guidance reward accordingly. A high guidance reward is given when the agent generates correct recommendations, which guarantees that the agent quickly and accurately comprehends the preferences of users. Finally, we jointly train the Actor-Critic network and the discriminator, when the whole training is done, the reasoning paths are taken as the interpretability of the recommendations. Extensive experiments on both the Scratch data set and public data set show that, Scratch-RL obtains favorable recommendation results compared with the state-of-the-art models.

With the rapid increase of users, Scratch, as a popular online social and programming platform, has accumulated massive project resources and complex relations across its social and programming learning network. However, it is challenging to utilize the network information for providing Scratch users with personalized services, despite there are already some useful network representation learning models. In this paper, a network representation learning model with preference-based generative adversarial nets for Scratch (ScratchGAN) is proposed to resolve this problem. In ScratchGAN, we first design a node-vector initialization approach to preserve structure information and side information of Scratch network. Then, considering to learn the fine-grained user preference information of network, we propose a novel Scratch adversarial learning model which includes a Scratch generative adversarial net and a user preference difference constraint component. The former aims to capture user preferences through a new generating strategy based on the delivery nature of preference. The latter attempts to embed users' detailed preference differences according to their interaction behaviors. ScratchGAN can mine user preferences while preserving network structure information and side information. Extensive experiments on the Scratch network show that ScratchGAN outperforms other state-of-the-art models in link prediction and recommendation tasks.

Applications consisting of a group of modular tasks can be offloaded to the multiaccess edge computing (MEC) for lower delay and energy consumption. In a dynamic MEC system, the fine-grained cooperative and dynamic offloading solution is necessary for the scenario of reusing tasks among devices. Considering the transmission cooperation, shared wireless bandwidth, and changing task queues on devices and edge servers, we formulate a joint offloading optimization problem to minimize the long-term average task execution cost. Although deep reinforcement learning (DRL) is a popular method for the dynamic problem, existing DRL algorithms are not suitable for our problem because of the hybrid discrete-continuous action spaces and constraints among action dimensions. Therefore, we propose a hybrid average reward proximal policy optimization (hybrid-ARPPO) algorithm to jointly optimize the offloading decisions, cooperative transmission ratios, and edge server assignments. First, we decompose our offloading problem into two subproblems. One is a tractable linear programming problem for continuous transmission ratios, and the other is a Markov decision process (MDP) only with discrete actions for offloading decisions and server assignments. Second, we take the expected average reward as the performance measure and deprecate the discount factor, which can reduce the work of tuning algorithms. Third, we design an action mask layer in the policy network of hybrid-ARPPO to filter invalid actions. Extensive experiments show the effectiveness of our hybrid-ARPPO in different system scales and task arrival patterns.

Mobile-edge computing (MEC) has been a promising solution for Internet-of-Things (IoT) applications to obtain latency reduction and energy savings. In view of the loosely coupled application, multiple devices can use the same task code and different input parameters to obtain diverse results. This motivates us to study the cooperation between devices for eliminating the repeated data transmission. Leveraging coalitional game theory, we formalize the cooperative offloading process of a reusable task into a coalitional game to maximize the cost savings. In particular, we first propose an efficient coalitional game-based cooperative offloading (CGCO) algorithm for the single-task model, and then expand it into a CGCO-M algorithm for the multiple-task model with jointly applying a two-stage flow shop scheduling approach, which helps to obtain an optimal task schedule. It is proved that our CGCO and CGCO-M can achieve the Nash-stable solution with convergence guarantee, and CGCO can obtain an optimal solution. The simulations show that CGCO is equal to the optimal exhaustive search (ES) method and CGCO-M is close to ES in terms of cost ratios. Cost ratios of CGCO and CGCO-M are significantly down by 41.08% and 83.70% compared to local executions, respectively. Meanwhile, CGCO-M obtains 41.46% and 89.74% reductions when reuse factors are 0.1 and 1, which means CGCO-M can save more cost with higher reuse density.

The Software Defined Network in Internet of Things (SDN-IoT) is enjoying growing popularity due to its flexibility, automaticity and programmability. However, there is still a lack of proper permission management on SDN-IoT applications (SIApps), especially when the SIApp’s required northbound interfaces are located in multiple heterogeneous controllers without mutual trust. Existing access control methods are usually based on centralized models, proprietary controllers, trusting conditions or manual operations. It can incur unnecessary performance degradation and poor scalability. To solve this problem, this paper proposes a SIApps’ ledger (SILedger), an open, trusted, and decentralized access control mechanism based on blockchain and attribute-based encryption (ABE). It can not only support effective authorization of SIApps in heterogeneous and untrusted SDN-IoT control domains, but also record all interactions between SIApps and resources, and thus facilitate SIApps further charging, analysis and audit. The main idea is that the SIApps are authorized using access tokens encrypted by ABE, and these tokens are seen as the currency of blockchain to be distributed. Specifically, we re-design blockchain transaction, token encryption, token initialization and token update schemes to achieve cross-domain, fine-grained and flexible SIApps’ permission management. In order to mitigate the delay and complexity problem of blockchain and ABE, we devise an access control framework that separates authorization from call process of SIApps. Finally, we perform security analysis and implement a FISCO-BCOS-based prototype for SILedger. The experimental results show that it can provide effective access control for SIApps with negligible overheads.

The mobile edge computing greatly reduces the transmission delay by offloading computing job to the base station, which is closer to the user. It solves the problem of high delay in cloud computing. Since the edge server has limited resources, improving the efficiency of computation offloading becomes a problem that needs to be addressed. Considering the data is reusable, it is a feasible solution to use the cache to improve the efficiency of computation offloading. In order to make the cache better serve the computation offloading process, we must design a reasonable cache replacement strategy. The traditional cache replacement strategy only considers the cache hit ratio, but ignores the cache value. It is not applicable to the edge computing process. In this paper, a new cache replacement strategy is proposed for the cache replacement problem in mobile edge computing, which is more suitable for computation offloading process. First, we define the cache value based on the priority of computing job. Then, we define the problem of maximizing the cache value in mobile edge computing as a multiobjective optimization problem. Finally, in order to find a suitable Pareto solution, we propose a cache replacement strategy based on the ideal point method. The experimental results show that the proposed algorithm results are significantly better than the existing competitive cache replacement methods.

Nowadays mobile multiplayer games have been an usual recreation, however energy-limited mobile devices confine the game running time. Noticing the component-based multiplayer game is a loosely coupled application whose components are nearly always required among players in one or more rounds, we study the collaboration of players to develop an energy-efficient offloading scheme. We formulate a 0-1 integer nonlinear programming problem to minimize the overall energy cost on player side under time-delay constraint. The problem is intractable to find the optimal solution, thus we propose a heuristic Two-phase Greedy-based Collaborative Offloading Algorithm (TGCOA) for the objective of cost minimization, while including more energy savings for low-energy players. In each round, we preferentially cache components saving more energy per data size for subsequent rounds, then preferentially offload components saving more energy for current round. Meanwhile, we always assign valid player with highest remaining energy to the uploading tasks. Simulations show that the energy cost ratios of our proposal are significantly down by 1.55% to 99.62% compared to four competing methods under different cache limitations and repeatability factors. Meanwhile, our proposal enables a 6.00% to 14.80% lower energy cost proportion for low-energy players compared to the four methods.

Review spam filtering of Chinese is a hot research topic in the field of natural language processing. In recent years, there have aroused a lot of neural network models for review spam filtering of Chinese, but these models mainly focus on utilizing text semantics or part-of-speech of a review without considering the variants of a sensitive word. In this paper, based on TextCNN, we propose a novel review spam filtering model of Chinese that embeds various features of the text. Considering the common types of variant words, we first extract three sets of vectors for each Chinese character of the review, including the single character vector, the pinyin vector, and the pinyin vector of each single character. Then, we use TextCNN to learn the features of these vectors, respectively. After that, the learned features are added as the comprehensive features of review. Finally, we input it to the softmax layer to get the final classification results. The experimental results show that the joint model performs better than classic classification methods in review spam filtering of Chinese. The recall and F1-score on review spam reach 92.4% and 93.6%.

Scratch, a visual programming language used by youth, has received widespread attention of education field. Quality Hound is an effective tool to detect the features of Scratch. However, its detection rules are not sufficiently complete, which incurs incomprehensive results. In this paper, we propose an ANTLR-based feature extraction and detection system to solve this problem. Specifically, nine novel programming feature detection rules are abstracted and applied in our model. The experimental results show our system can effectively extract programming features from projects and provide feedback for students and teachers.

Software defined networks have been a driving force of the Internet of Things (IoT) advancement in the devices management and network control. One of the important challenges remaining to be resolved is how to safely complete the network update with low resources' consumption. In this paper, we propose a new batch-level update mechanism with low cost (BLLC) to solve this problem. Its main idea is to bundle control commands and update networks from the destination to the source of new flows. Specifically, we first build update trees for all new flows according to the network states and consistency properties. Then, the new rules in an update tree are packaged into a single control packet (UBCP). Furthermore, a virtual destination-based algorithm is devised to minimize the number of finally formed UBCPs. The generated UBCPs will successively pass all IoT nodes to be updated, and instruct them to apply the required operations. If the control channel is wireless, we introduce a cooperative transmission mechanism to enhance the reliability of the UBCPs' forwarding. Finally, we evaluate our scheme by comprehensive experiments. The results show that the BLLC reduces the link cost by 68% on the average with a slight performance loss in the update time.

One of the crucial technologies for future Internet of Things (IoT) is software defined networks, which provides a centralized and programmable control ability for operators. However, the current single control plane deployed on the remote IoT gateway may incur a bottleneck with the continuous growth of IoT devices and applications. In this paper, we propose a two-level hierarchy (the master and slave) control framework to resolve this problem. Additionally, a novel slave controller placement strategy (SCPS) is presented to further optimize the control performance. In SCPS, we first design a synthetic IoT node importance assessment model based on an improved analytic hierarchy process and fuzzy integral. It considers the device attributes, service attributes, and control frequency. Then we formulate the slave controller placement as a binary integer program problem. It is solved by a modified binary particle swarm optimization algorithm to optimize the control delay and control cost of critical IoT nodes. Finally, we carry out extensive experiments to evaluate the performance of our strategy. The results show that, compared to other competing methods, it approximately reduces the control delay of important IoT nodes by 30.56%.

Named Data Network (NDN) can cache data in the router to reduce the data transmission cost and hence can be used in the resource limited Internet of Things systems. However, there are still some problems. Particularly for periodic data, if the cached data can't be updated in real time, the data content will be outdated. However, if each cache node updates the data frequently, it will cost a lot of unnecessary network traffic. In this paper, we improve the traditional NDN routing procedure and propose an active updating strategy for periodic data to further reduce the network traffic. First, we create an Interest Record Table and modify the Content Store to store the periodic information. Then, we modify the router work-flow to allow the router node to actively update the cache data. Finally, we propose an algorithm based on user's access regularity to calculate the update time point. Experimental results indicate that the proposed approach can reduce the network traffic by about 33.8% on average compared to the existing competing methods.

In the context of mobile crowd sensing, incentive mechanism is crucial to recruit mobile users to participate in the sensing task and ensure participants to provide high-quality sensing data. In this paper, we investigate a staged incentive mechanism for mobile crowd sensing. We firstly divide the incentive process into two stages: recruiting stage and sensing stage. In the recruiting stage, we introduce the payment incentive coefficient and design a Stackelberg based game method. The participants can be recruited via game interaction. In the sensing stage, we propose a time-space correlation algorithm in the interaction and the winners can be screened after the sensing task. Finally, Extensive experiments show that compared to the existing positive auction incentive mechanism (PAIM) and reverse auction incentive mechanism (RAIM), our staged incentive mechanism (SIM) can effectively improve participants' motivation and achieve high-quality sensing data from both space dimension and time dimension by extending the motivation from the recruitment stage to the sensing process.

E-Health Systems are transforming the whole healthcare process to become more efficient and less expensive. QoS is usually used as a key criterion for E-Health service. The study shows that QoS of data supply chain are highly related to operating mode of nodes. However, existing optimization methods seldom took this observation into account, which shall decrease the optimal performance. In this paper, we propose an operating mode optimization strategy for nodes of data supply chain environments including E-Health Systems. First, a QoS mathematical model is developed. Then, we put forward the objective function to balance the data freshness, hit rate and cost from the aspects of users and service providers. To this end, we present a novel operating mode selection algorithm. It selects appropriate candidate operating modes for each node when generating composite operating mode of nodes with optimal QoS values. Experimental results indicate that the effectiveness of the proposed approach outperforms the existing methods by at least 20%.

Due to the execution paradigm may be different at different invocation time, users obtain different QoS when interacting with the same Data Supply Chain (DSC). However, existing QoS prediction methods seldom took this observation into consideration, which shall decrease the prediction accuracy. In this paper, we propose a context-based QoS prediction method for data supply chain. First, a QoS mathematical model is developed for considering the mass data transmission across elementary sub-chains. Then, two execution paradigms of data supply chain are discussed. Besides, we explored several special context factors of data supply chain (such as invocation time, data source update period and execution paradigm) which influence QoS. By processing such context information, we can obtain the part of data supply chain which is need to execute when the user query occurs and leverage them to predict QoS. Experimental results indicate that our approach improves the prediction accuracy and efficiency of QoS when compared to previous methods.

With the development of social network, there are more and more people to share pictures on social platforms. Since the information contained in picture has the different requirements for confidentiality, it makes the selective presentation of secret information to be an urgent problem. Estimating user's privilege of gaining some regions based on his/her attributes is a novel solution. But there are few perfect solutions aiming at the strategy of adaptive calculation for user's privilege in the existing literatures, especially for the scenario in which the real values of some attributes have priorities. In this work, based on the cipher text-policy attribute-based encryption (CP-ABE), we propose an adaptive and safe presenting scheme for the information contained in a picture. This scheme firstly embeds the confidential data outside the secret region, and generates the image mosaic in the secret region; when someone requesting the original version of image, it adaptively calculates the recovery privilege of requestor with the strategy proposed in this paper, then precisely present some regions based on the privilege level of requestor. Moreover, we firstly propose the vote-attribute which facilitates the attribute revocation. The experiments demonstrate that, based on the privilege level, the proposed scheme can safely present the original version of the corresponding image region, and expediently achieve the attribute revocation. Compared with other algorithms, our scheme can restore the original version of image with only 1/2 secret data, and spend little time over the attribute revocation. Besides, the average of peak signal to noise ratio (PSNR) is 4dB more than the algorithms available, and the standard variance of PSNR is less than 0.4.

It is common to take photos and share them by smartphone in mobile social network at present, but the sharer may need to mask partial areas of the photo to avoid leaking the privacy. Reversible image mosaic can mask the privacy very well. However, there is little discussion on how to effectively control and safely store the mosaic information for recovering. In this paper, we first propose a prediction algorithm of reversible image mosaic size (PARIMoS), which can calculate the maximum number of mosaic blocks according to the image quality of non-mosaic area and the recovery quality of mosaic required by user. And then, we propose an efficient and reliable generating algorithm of reversible image mosaic (GAoRIM). In GAoRIM, we compress the secret information of the mosaic area with twice-run-length-encoding and then store the compressed result in special non-mosaic areas of image separately. Therefore, it can ensure reliable recovery even the image is partly stained, and hardly change the roughness of non-mosaic area. The experiment results show that the biggest change of average roughness in the non-mosaic area is less than 0:5%. Comparing with the similar algorithms, on average, the amount of data protected by GAoRIM is at least 8:70% higher than theirs while the peak signal noise ratio (PSNR) of stego image without mosaic is similar to theirs.

Indoor localization is an important primitive that can enable many ubiquitous computing applications. This paper improves the scheme of landmark and inertial navigation through crowd sensing. The location of landmark can calibrate the position of users, and accurate localization can optimize the landmark's location in turn. In this work, we define landmarks as certain characteristic structures with iBeacons. To tackle the challenge of low efficiency of a landmark, we combine Bluetooth signals and sensor readings to reflect the distinctive signature of a landmark and then design a method for a reliable detection of a landmark. Moreover, we investigate the calibration bias of a landmark to ensure the calibration accuracy. For improving the accuracy of inertial navigation, we personalize the step length model and correct the user's heading with the aid of crowd sensing. We have built an indoor localization system integrating the above modules and an indoor floor map, which can be further improved with more users using our system. We demonstrate for the first time a meter-level indoor localization system that is self-improving, user adaptive, and easy to deploy. Extensive experiments on users with mobile devices, with over 37 subjects walking over an aggregate distance of over 30 km, were carried out. Evaluation results show that our system can achieve a mean accuracy of 2 m initially and 1 m with the calibration of landmarks in a 39 m × 21 m testing area. Copyright © 2016 John Wiley & Sons, Ltd.

There are more and more systems using mobile devices to perform sensing tasks, but these increase the risk of leakage of personal privacy and data. Data hiding is one of the important ways for information security. Even though many data hiding algorithms have worked on providing more hiding capacity or higher PSNR, there are few algorithms that can control PSNR effectively while ensuring hiding capacity. In this paper, with controllable PSNR based on LSBs substitution- PSNR-Controllable Data Hiding (PCDH), we first propose a novel encoding plan for data hiding. In PCDH, we use the remainder algorithm to calculate the hidden information, and hide the secret information in the last x LSBs of every pixel. Theoretical proof shows that this method can control the variation of stego image from cover image, and control PSNR by adjusting parameters in the remainder calculation. Then, we design the encoding and decoding algorithms with low computation complexity. Experimental results show that PCDH can control the PSNR in a given range while ensuring high hiding capacity. In addition, it can resist well some steganalysis. Compared to other algorithms, PCDH achieves better tradeoff among PSNR, hiding capacity, and computation complexity.

In this paper, we propose a habit-based SWRL generation and reasoning approach in smart home. Definition and recognition of habits of daily living can provide humanized smart home for assisted living application, especially for people with memory deficits. This paper presents Recognizing Habit of Daily Living(RHDL) by discovering and monitoring smart home context information. The habit and habit association of using electrical appliances are defined explicitly for the first time. The generation rules between habit/complex habit and SWRL are designed, and the reasoning is based on the Semantic Web Rule Language(SWRL). The ontology model for the RHDL is designed and the prototype system of RHDL is implemented using protege and Jess tools.

Smart building systems can provide flexible and configurational sensing and controlling operations according to users' requirements. As the number and the complexity of service rules customized by users have significantly increased, there is an increasing danger of conflict during the interaction process between users and the system. To address this issue, we propose a new rule conflict detection scheme tailored for the smart building system. First, we present a formal rule model UTEA based on User, Triggers, Environment entities, and Actuators. This model can handle not only controlled devices with discrete status but also real-valued environmental data such as temperature and humidity. In addition, this model takes multiple users with different authorities into account. Second, we define 11 rule relations and further classify conflicts into five categories. Third, we implement a rule storage system for detecting conflicts and design a conflict detection algorithm, which can detect the conflict between two rules as well as cycle conflict/multicross contradiction among multiple rules. We evaluated our scheme in a real smart building system with more than 30 000 service rules. The experiment results show that our scheme improves the performance in terms of error/missed-detection rates and running time.

Clustering sensor nodes is an effective method for achieving high energy efficiency and flexible data aggregation in wireless sensor networks. However, scheduling all clusters to avoid the inter-cluster interference and meanwhile send all observation to the sink quickly is a difficult problem. In this paper, we use interference graph to model the inter-cluster collision problem and propose an efficient sleep scheduling mechanism in the clustered wireless sensor networks. We firstly divide the clusters into layers, then transform the cluster-tree network into the interference graph of clusters, and thereby divide all clusters on one layer into different independent sets. All clusters in the same independent set can work simultaneously without interference, while all independent sets are scheduled one by one for collision-free. To reduce the total delay of data report in the whole network, we propose three kinds of rules to sort the independent sets. Simulation and analysis show that the degree-based sorting rule works best. The practical experiments also verify this conclusion.

In Wireless Multimedia Sensor Networks (WMSNs), data fusion and collaborative in-network processing operations often require effective multimedia synchronization control. Extensive researches have been done in the traditional networks. Most of these works assume that there exists a powerful synchronization controller in the network. However, for WMSNs, the in-network processing of the multimedia content is usually performed based on the resource-constrained sensors. Traditional synchronization control algorithms fail to run on the low-end hardware platform. In this paper, we propose an adaptive synchronization control scheme with multi-level buffer to address this problem. Main contributions of this paper are as follows. (i) To overcome the resource constraint of WMSNs, we perform the synchronization control in a distributed way by combining the sensors in the stream transmission path. (ii) By evaluating the current network state, we adjust the level number of the buffer adaptively to balance the delay and delay jitter. (iii) Through analyzing the buffer state of each level, we propose an efficient packet scheduling scheme to reduce the delay jitter actively. (iv) We implement the proposed scheme and verify its effectiveness in our practical WMSNs platform. The experiment results show that the proposed scheme can maintain the streams synchronization effectively with the low-end sensors.

For wireless multimedia sensor networks (WMSNs) deployed in noisy and unattended environments, it is necessary to establish a comprehensive framework that protects the accuracy of the gathered multimedia information. In this paper, we jointly consider data aggregation, information trust, and fault tolerance to enhance the correctness and trustworthiness of collected information. Based on the multilayer aggregation architecture of WMSNs, we design a trust-based framework for data aggregation with fault tolerance with a goal to reduce the impact of erroneous data and provide measurable trustworthiness for aggregated results. By extracting statistical characteristics from different sources and extending Josang's trust model, we propose how to compute self-data trust opinion, peer node trust opinion, and peer data trust opinion. According to the trust transfer and trust combination rules designed in our framework, we derive the trust opinion of the sink node on the final aggregated result. In particular, this framework can evaluate both discrete data and continuous media streams in WMSNs through a uniform mechanism. Results obtained from both simulation study and experiments on a real WMSN testbed demonstrate the validity and efficiency of our framework, which can significantly improve the quality of multimedia information as well as more precisely evaluate the trustworthiness of collected information.

Wireless Sensor Networks are attracting lots of attention for its broad application areas, and hence the services provided by WSNs become much more diverse. But with the development of in-depth applications, service discovery and composition are challenges to the end users. Service mining is a new technique in Internet which can discover new composite services from the abundantly existing services in Internet to meet the users' needs. However, those service mining methods are not practical in the energy-exhausted and application-oriented WSNs. In this paper, we develop a semantics-based service mining method for WSNs, to provide users with interesting composite services. In this method, services are combined and recommended to users actively according to the calculation of service similarity and an up datable semantic database. By the results of service similarity, useless compositions are filtered out so that energy consumption on service flooding can be reduced. The update strategy of semantic database is also given out, by which the composite services can keep up with time and be more applicative. The benefits of the proposed method are that all operations such as calculating, filtering, and updating are simple enough and can be performed at broker nodes.

In Wireless Audio Sensor Networks (WASNs), many applications such as target tracking, data fusion and localization, often require precise synchronization among audio streams. Extensive researches have been done under the assumption that wireless channel is well enough and the cluster head is power supplied. However, these conditions can not be satisfied in many applications. When audio event occurs, the burst of data is usually coupled with a high packet loss rate. Besides, the cluster head is usually battery-powered, especially in the wild environment. In this paper, we propose a distributed audio synchronization scheme to address these challenges. By dynamically detecting the endpoint of the audio streams and marking the start point of the audio synchronization obviously at the reporting sensor, we can eliminate the sound propagation delay effectively and synchronize the audio streams precisely even under the high packet loss rate. Moreover, by handling the synchronization task in a distributed way, we can lighten the burden of the cluster head and prolong the lifetime of the whole network greatly. To improve the flexibility of the scheme, we also propose a self-adaptive algorithm which can adjust the parameters of the synchronization according to the network condition. Experimentally, we show that the proposed scheme can synchronize the audio streams under different conditions with low energy cost.

Video surveillance system, which can provide real-time display of the monitored scene and video playback, has been employed in many areas including: commercial security, accident investigation, law enforcement and emergency response. However, audio which carries important information not available in video is usually not taken seriously and used effectively. In this paper, we develop an enhanced surveillance platform by introducing the low-power wireless audio sensor networks (WASNs). We can obtain more comprehensive and precise monitoring without the limitation of the line-of-sight and lighting condition. Moreover, this platform is designed and built for providing key support to varieties of applications. This article describes the platform architecture, including design, implementation, and performance. We describe the audio sensor platform which can deliver high-quality audio over sensor network by multi-hops with low power requirement. In addition, we present the multimedia synchronization mechanism in the heterogeneous network which is the foundation of applications in the proposed platform. Our experiments include an in-depth analysis of the bottlenecks within the platform as well as measurements for the various components.

Energy-efficient and reliable transmission of sensory information is a key problem in wireless sensor networks. To save more energy, in-network processing such as data fusion is a widely used technique, which, however, may often lead to unbalanced information among nodes in the data fusion tree. Traditional schemes aim at providing reliable transmission to individual data packets from source node to the sink, but seldom offer the desired reliability to a data fusion tree. In this paper, we explore the problem of Minimum Energy Reliable Information Gathering (MERIG) when performing data fusion. By adaptively using redundant transmission on fusion routes without acknowledgments, packets with more information are delivered with higher reliability. For different data fusion topologies, such as star, chain, and tree, we provide optimal solutions to compute the number of transmissions for each node. We also propose practical, distributed approximation algorithms for chain and tree topologies. Analytical proofs and simulation results show that energy-efficient information reliability can be guaranteed in an unreliable wireless environment with the help of our proposed schemes.

Wireless Audio Sensor Networks (WASNs) can provide event detection, object tracking and audio stream monitoring through cooperative audio sensor nodes. Extensive researches have focused on sound detection and source localization, but little work is on the monitoring and recovery of audio stream. To clearly reconstruct the real-time audio stream from WASNs, both the quantity and quality of samples provided by sensor nodes should be carefully considered. In this work, we propose a novel approach termed "adaptive sampling and diversity reception", which can significantly reduce the energy consumption of network while maintaining high accuracy of audio stream. On one hand, by selectively picking sampling nodes and adaptively adjusting their sampling rates, sufficient samples arrive at the fusion node for high recovering fidelity. On the other hand, through Maximal-Ratio-Combining(MRC)-like signal fusion scheme at the fusion node, the signal-to-noise ratio (SNR) of combined audio is further improved. Meanwhile, we propose a heuristic algorithm of sampling rate assignment which can select a set of optimal sampling nodes and hence minimize the energy consumption on data transmission. Analytically and experimentally, we show that the proposed scheme can improve the signal recovering fidelity tremendously with low energy cost.

Wireless Audio Sensor Networks (WASNs) can provide event detection, object tracking and emergency response through cooperative audio sensor nodes. Effective collaboration of audio sensors requires precise synchronization among audio streams. Some researches have been done on the timestamp mechanism based on time synchronization ignoring propagation delay and many other researches have focused on the synchronization of simple gunshot or scream. However, for the synchronization of intermittent and fluctuating audio stream, there still exists many challenges. In this paper, we propose an effective audio synchronization scheme which can synchronize the intermittent audio streams adaptively while maintain low energy cost. On one hand, we obtain audio synchronization without global clock which save energy tremendously. On the other hand, by introducing a feedback loop mechanism, we can keep a high audio synchronization fidelity even when the audio source moves around and the sound strength varies with time. Furthermore, we discuss the extension for flexibility and scalability of this scheme when there exist several sound sources simultaneously or the audio source moves among clusters. Through experiments on a WASNs platform and simulations, we show that the proposed scheme is desirable to guarantee the accuracy of audio synchronization in practical environment with low energy cost.

In sensor networks, en route aggregation decision regarding where and when aggregation shall be performed along the routes has been explicitly or implicitly studied extensively. However, existing solutions have omitted one key dimension in the optimization space, namely, the aggregation cost. In this paper, focusing on optimizing over both transmission and aggregation costs, we develop an online algorithm capable of dynamically adjusting the route structure when sensor nodes join or leave the network. Furthermore, by only performing such reconstructions locally and maximally preserving existing routing structure, we show that the online algorithm can be readily implemented in real networks in a distributed manner requiring only localized information. Analytically and experimentally, we show that the online algorithm promises extremely small performance deviation from the offline version, which has already been shown to outperform other routing schemes with static aggregation decision.

While in-network data fusion can reduce data redundancy and, hence, curtail network load, the fusion process itself may introduce significant energy consumption for emerging wireless sensor networks with vectorial data and/or security requirements. Therefore, fusion-driven routing protocols for sensor networks cannot optimize over communication cost only—fusion cost must also be accounted for. In our prior work [2], while a randomized algorithm termed MFST is devised toward this end, it assumes that fusion shall be performed at any intersection node whenever data streams encounter. In this paper, we design a novel routing algorithm, called Adaptive Fusion Steiner Tree (AFST), for energy efficient data gathering. Not only does AFST jointly optimize over the costs for both data transmission and fusion, but also AFST evaluates the benefit and cost of data fusion along information routes and adaptively adjusts whether fusion shall be performed at a particular node. Analytically and experimentally, we show that AFST achieves better performance than existing algorithms, including SLT, SPT, and MFST.

In this paper, we propose a routing algorithm called Minimum Fusion Steiner Tree (MFST) for energy efficient data gathering with aggregation (fusion) in wireless sensor networks. Different from existing schemes, MFST not only optimizes over the data transmission cost, but also incorporates the cost for data fusion, which can be significant for emerging sensor networks with vectorial data and/or security requirements. By employing a randomized algorithm that allows fusion points to be chosen according to the nodes' data amounts, MFST achieves an approximation ratio of {\frac{5}{4}}\log(k+1), where k denotes the number of source nodes, to the optimal solution for extremely general system setups, provided that fusion cost and data aggregation are nondecreasing against the total input data. Consequently, in contrast to algorithms that only excel in full or nonaggregation scenarios without considering fusion cost, MFST can thrive in a wide range of applications.