Search

With the emergence of a large-scale pre-training model based on the transformer model, the effect of all-natural language processing tasks has been pushed to a new level. However, due to the high complexity of the transformer's self-attention mechanism, these models have poor processing ability for long text. Aiming at solving this problem, a long text processing method named HBert based on Bert and hierarchical attention neural network is proposed. Firstly, the long text is divided into multiple sentences whose vectors are obtained through the word encoder composed of Bert and the word attention layer. And the article vector is obtained through the sentence encoder that is composed of transformer and sentence attention. Then the article vector is used to complete the subsequent tasks. The experimental results show that the proposed HBert method achieves good results in text classification and QA tasks. The F1 value is 95.7% in longer text classification tasks and 75.2% in QA tasks, which are better than the state-of-the-art model longformer.

Continuous data scale growth increases energy consumption and operating cost that cannot be ignored in cloud storage systems. Previous studies have shown that analyzing the characteristics of I/O access and mining data features is effective for reasonable data distribution in storage systems. The granularity and criterion of classification are the key factors in determining the data distribution. To decrease energy consumption and operating cost, this paper puts forward a fine-grained framework of the climatic-season-based energy-aware in cloud storage system called CSEA. The framework concludes the following three aspects: (i) data feature mining. CSEA discovers potential data features by analyzing data access to provide help with data classification. (ii) K-means clustering algorithm. CSEA uses an unsupervised data classification algorithm in machine learning to divide data into categories based on seasonal characteristics by gathering real I/O access. (iii) data distribution of fine-grained. On the basis of seasonal features, CSEA fuses regional features to further refine the data distribution granularity to save on energy consumption and operating cost. Simulation experiments using extended CloudSimDisk and the constructed mathematical models indicate that CSEA reduces the energy consumption and operating cost compared with the single data classification standard and coarse-grained data distribution.

Thanks to excellent reliability, availability, flexibility and scalability, redundant arrays of independent (or inexpensive) disks (RAID) are widely deployed in large-scale data centers. RAID scaling effectively relieves the storage pressure of the data center and increases both the capacity and I/O parallelism of storage systems. To regain load balancing among all disks including old and new, some data usually are migrated from old disks to new disks. Owing to unique parity layouts of erasure codes, traditional scaling approaches may incur high migration overhead on RAID-6 scaling. This paper proposes an efficient approach based Short-Code for RAID-6 scaling. The approach exhibits three salient features: first, SS6 introduces $\tau $ to determine where new disks should be inserted. Second, SS6 minimizes migration overhead by delineating migration areas. Third, SS6 reduces the XOR calculation cost by optimizing parity update. The numerical results and experiment results demonstrate that (i) SS6 reduces the amount of data migration and improves the scaling performance compared with Round-Robin and Semi-RR under offline, (ii) SS6 decreases the total scaling time against Round-Robin and Semi-RR under two real-world I/O workloads (iii) the user average response time of SS6 is better than the other two approaches during scaling and after scaling.

The overload of data always threatens the reliability of storage system. The RAID-6 storage system provides higher reliability and flexible scalability. RAID-6 scaling can rapidly relieve the insufficient storage capacity in a short time. Therefore, this paper proposes Horizontal Data migration Scaling (HDS), an efficient RAID-6 scaling scheme, for HDP Code. First, it only migrates a small amount of data from the old disk to the new disk to regain I/O load balancing among all disks including old and new. Second, it optimizes the update order of anti-diagonal parity data to reduce the cost of parity data update. By numerical results and real experimental data analysis, this paper compares the performance of HDS to Round-Robin and Semi-RR. Compared with Round-Robin and Semi-RR, the final analysis results indicate: (1) HDS reduces the data migration by 59.9 $$\sim$$ 83.3%; (2) HDS decreases the total cost of XOR operations by 36.84 $$\sim$$ 71.43% and 66.04 $$\sim$$ 76.92%; (3) HDS improves the total scaling time by 43.78 $$\sim$$ 61.83% and 16.39 $$\sim$$ 48.89% under offline.

Caused by massive data, large-scale data centers suffer from insufficient storage capacity. This increases the risk of component damage and disk failure. To enhance storage capacity of data centers, scaling has turned out to be one of the most popular choices. RAID-6 is deployed extensively that has superior reliability, availability, and scalability compared to other RAID levels. However, higher reliability also means expensive parity data update cost. To address this issue, this paper proposes an efficient RAID-6 scaling scheme, HS6, for H-Code. The basic idea of HS6 is reducing the total parity data update cost and saving the total scaling time by optimizing data migration and parity data update. The properties of HS6 can be summarized as follows: (1) it utilizes horizontal data migration to eliminate the horizontal parity data update, and (2) it makes full use of original parity data to cut down the total parity data update cost. Numerical results and experimental data analysis indicate that: (1) HS6 decreases the data migration rate by 76.92% $$\sim$$ 95.31%, 55.56% $$\sim$$ 84.46%, and 14.29% $$\sim$$ 16.67% compared to Round-Robin, Semi-RR, and HCS, (2) HS6 saves the total scaling time by 62.05% $$\sim$$ 76.75%, 53.24% $$\sim$$ 73.94%, and 3.68% $$\sim$$ 6.79% against Round-Robin, Semi-RR, and HCS, respectively, under offline and outperforms Round-Robin by 59.99% $$\sim$$ 79.02%, Semi-RR by 53.04% $$\sim$$ 76.41%, and HCS by 0.56% $$\sim$$ 23.78% in total scaling time under four workloads, and (3) HS6 maintains almost the same user average response time with Round-Robin, Semi-RR, and HCS during scaling and after scaling.

An elastic stream computing system needs elastic adjustment of computing resource allocation and vertex parallelism to improve latency and throughput, which includes continuously or periodically scaling in/out the workload of computing nodes at runtime. Dynamic redirection can help with this elasticity issue by dynamically redirecting real-time data streams to computing resources. Due to the time-varying and unpredictable nature of real-time data streams, implementing redirection of data streams is challenging. Currently, the requirements of data streams redirection are not fully fulfilled, which directly affects the latency and throughput of stream computing systems. To bridge this gap, we proposed a dynamic redirection framework (called Dr-Stream) for elastic stream computing systems. This paper discussed the following aspects: (1) Investigating the dynamic redirection of real-time data streams, providing a general stream application model, a data stream model and a data stream grouping model, as well as formalizing the problem of load balancing optimization and data stream redirection. (2) Redirecting data streams among multiple instances of an operator at runtime by a lightweight load balancing strategy to improve the load balancing of a data center at the vertex level. Managing system states, especially the states of stateful operators by a logical ring-based strategy to improve accuracy. (3) Determining the number of instances for each operator, and deploying the instance(s) to computing nodes by a modified first-fit strategy at runtime. (4) Evaluating the fulfillment of low latency, high throughput, and load balancing objectives in a real-world distributed stream computing environment. Experimental results showed that Dr-Stream reduced the average system latency and load balancing of the data center by more than 20% and 15%, respectively. It also improved the average system stability by more than 15% and avoided over-utilization of computing nodes, as compared to the existing strategies in Storm.

At present, the mainstream distant supervised relation extraction methods existed problems: the coarse granularity for coding the context feature information; the difficulty in capturing the long-term dependency in the sentence, and the difficulty in coding prior knowledge of structures are major issues. To address these problems, we propose a distant supervised relation extraction model via DiSAN-2CNN on feature level, in which multi-dimension self-attention mechanism is utilized to encode the features of the words and DiSAN-2CNN is used to encode the sentence to obtain the long-term dependency, the prior knowledge of the structure, the time sequence, and the entity dependence in the sentence. Experiments conducted on the NYT-Freebase benchmark dataset demonstrate that the proposed DiSAN-2CNN on a feature level model achieves better performance than the current two state-of-art distant supervised relation extraction models PCNN+ATT and ResCNN-9, and it has d generalization ability with the least artificial feature engineering.

During the past ten years, the energy consumption problem in cloud-related environments has attracted substantial attention in research and industrial communities. Researchers have conducted many surveys on energy efficiency issues from different perspectives. All of the surveys can be classified into five categories: surveys on the energy efficiency of the whole cloud related system, surveys on the energy efficiency of a certain level or component of the cloud, surveys on all of the energy efficient strategies, surveys on a certain energy efficiency techniques, and other energy efficiency related surveys. However, to the best of our knowledge, surveys on energy-aware data management strategies in cloud-related environment are absent. In this paper, we conduct a comprehensive survey on energy saving-aware data management strategies in cloud-related environments, such as data classification, data placement and data replication strategies. Compared to current existing reviews on energy efficiency in cloud-related environments, we firstly conduct the survey on the energy consumption problem from the data management perspective. Furthermore, we classify the energy-aware data management strategies from different perspectives. This survey and the taxonomy of the energy-aware data management strategies demonstrate the potential for reducing the energy consumption at the data management level of a cloud storage system, which will compress more space for energy reduction and finally achieve energy proportionality. Moreover, this survey and taxonomy on the energy efficiency issue from the data management perspective is an important supplement to current existing surveys on energy efficiency in cloud-related environments.

The recall of defective automobile products is one of the important measures to promote the quality of product quality and protect consumers' pyhsical safety and property security. In order to assess the risk level of defect cases, automobile recall management experts need to analyze and discuss the defect information by personal. A risk level prediction method based on language pre-training Bert model is proposed in this paper, which can transform the defect information into rick level of the vehicle and then predict vehicle recall automatically, in which a seq2seq model is proposed to multi-label the vehicle complaint data. The outputs of the seq2seq model combined with other static and dynamic information are used as the input of the Bert communication model. Substantial comparative experiments of different feature combinations on different methods show that the proposed VDRF method achieves F1 value with 79% in vehicle recall risk prediction, which outperforms the traditional method.

Although image captioning technology has made great progress in recent years, the quality of Chinese image description is far from enough. In this paper, we focus on the problem of Chinese image captioning with the aim to improve the quality of Chinese image description. A novel framework for Chinese image captioning based on image label information (CIC) is proposed in this paper. Firstly, image label information is extracted by a multi-layer model with shortcut connections. Then the label information is input into the neural network with an extension of LSTM, which we coin L-LSTM for short, to generate the Chinese image descriptions. Extensive experiments are conducted on various image caption datasets such as Flickr8k-cn, Flickr30 k-cn. The experimental results verify the effectiveness of the proposed framework (CIC). It obtains 27.1% and 21.2% BLEU4 average values of Flickr8k-cn and Flickr30k-cn, respectively, which outperforms the state-of-art model in Chinese image captioning domain.

As the country vigorously promotes the development of science and technology and tries to enhance independent innovation capabilities, more and more attention is paid on the protection of technology ownership. In recent years, China has developed rapidly in many scientific and technological fields, and the number of patent applications increased year by year. However, various patent quality problems including immature patent technology and low patent authorization rate appear. The indicators of patent quantification and quality evaluation are studied in this paper. First, we quantify the patent quality evaluation indicators and combine the content of the patent text to build a patent evaluation model. US patents with patent grade labels are used for training with multitask learning technology. Second, the evaluation model is transferred from the English patents to the Chinese patents, in which the active learning technology and transfer learning technology are used to minimize the work of manual labeling. Finally, a Chinese patent quality evaluation model based on collaborative training was designed and implemented. Methods used in this experiment have notably improved the prediction effect of the model and achieved a better migration effect. A large number of experimental results show that the Chinese patent quality evaluation model has achieved good evaluation results. This research uses deep learning and natural language processing technology to carry out research on patent quality evaluation models from different perspectives, to provide patent decision support for related companies, and to point out research directions for research institutions and patent inventors.

The effective organization and utilization of military equipment data is an important cornerstone for constructing knowledge system. Building a knowledge graph in the field of military equipment can effectively describe the relationship between entity and entity attribute information. Therefore, relevant personnel can obtain information quickly and accurately. Attribute extraction is an important part of building the knowledge graph. Given the lack of annotated data in the field of military equipment, we propose a new data annotation method, which adopts the idea of distant supervision to automatically build the attribute extraction dataset. We convert the attribute extraction task into a sequence annotation task. At the same time, we propose a RoBERTa-BiLSTM-CRF-SEL-based attribute extraction method. Firstly, a list of attribute name synonyms is constructed, then a corpus of military equipment attributes is obtained through automatic annotation of semistructured data in Baidu Encyclopedia. RoBERTa is used to obtain the vector encoding of the text. Then, input it into the entity boundary prediction layer to label the entity head and tail, and input the BiLSTM-CRF layer to predict the attribute label. The experimental results show that the proposed method can effectively perform attribute extraction in the military equipment domain. The F 1 value of the model reaches 77% on the constructed attribute extraction dataset, which outperforms the current state-of-art model.

Sentiment analysis of film reviews is the basis of obtaining the opinions of movie viewers. It has an important influence on movie public opinion control and stimulating potential viewers. Due to the natural openness and randomness of social media, there may exist a considerable amount of useless or false information in film review comments, making it challenging to analyze the credibility of the comments. This paper proposes a fine-grained sentiment analysis method based on the key-viewpoint sentences of Chinese film reviews, where a deep learning model is used to classify the fine-grained emotions in film reviews. Based on the analysis results, a method for calculating the credibility of review comments is proposed. Under the credibility criteria, corpus screened through credibility filtering algorithm, the overall sentiment classification can obtain 9% improvement on accuracy than the original corpus, which verifies the validity of the credibility algorithm. The higher quality corpus achieved by the credibility algorithm is benefit for improving the accuracy of the sentiment classification.

At present, China has become a major patent production country, and the number of patent applications has been ranked first in the world for many years. As the number of patents has increased, the quality of patents has begun to draw people’s attention. At present, there is no clear evaluation method for Chinese patents. Manual evaluation of patents requires a large number of relevant experts to research and compare patents in different fields, which is time-consuming and labor-intensive. In the previous study, the author constructed an English patent quality evaluation model PQE-MT using U.S. Patents that represent patent strength. This paper introduces this model into Chinese patents through transfer learning and active learning, thereby reducing the workload of manual labeling. The evaluation results show that the method in the experiment has achieved a good migration effect, with Micro-F1 reaching 74%.

Rapid increase in energy consumption is a serious problem in cloud storage systems. Data accessed in large-scale storage systems usually exhibit temporal and spatial characteristics, which make it possible to reduce energy consumption by clustering data with similar access characteristics for storage in the same zone of cloud storage systems. Existing works usually only focus on the frequency of data access. However, widely existing phenomena show data access with seasonal and tidal characteristics in cloud storage systems. The seasonal and tidal characteristics of data access are extracted thoroughly in this paper. According to the extracted data access characteristics, energy-aware data clustering through a machine learning algorithm (K-ear) is proposed. K-ear classifies data into five seasonal categories according to their seasonal access characteristics and then classifies every seasonal category into three tidal categories according to its tidal access characteristics. The 15 classified categories are stored in different storage zones with different energy and performance modes. Simulation experiments using CloudSimDisk with the constructed mathematic models demonstrate that the proposed K-ear algorithm is more energy-efficient than the default data clustering algorithms in Hadoop and the classical data clustering storage strategy according to the data access frequency (Striping-Based Energy-Aware Strategy).

This article describes how with the continuous expansion on the volume of data produced by sensors in Cyber Physical Systems, the scale of the cloud storage system has become larger. This will lead to the problems of a high energy consumption rate and a low utilization becoming a serious issue. In order to enhance the effective energy consumption, reduce the invalid energy consumption, and supply more flexible QoS for users in CPS, this article proposes an automatic energy gear-shifting mechanism with flexible QoS constraints (QGLG). The QGLG predicts system load of the follow-up period through a support vector machine model. According to the current system load, the predicted load, and the flexible QoS, QGLG automatically up-shifts and down-shifts among nodes. Substantive results from the simulation experiments done on GridSim show that the QGLG can achieve energy consumption reduction while satisfying the user's flexible QoS requirements. Compared with a similar energy-reducing mechanism, QGLG has its obvious advantage when considering the requirements of user with energy saved notwithstanding.

The problem of energy consumption in the cloud related environments has attracted a great deal of attention in research and industry communities. In the past ten years, a large number of studies have been done on energy efficiency algorithms, methods, strategies, or techniques. Over the past five years, many scholars have conducted surveys on those energy efficiency strategies from different perspectives. In this paper, we conduct a survey and build the taxonomy on current existing energy efficiency relevant surveys, that is, a survey on surveys of energy efficiency. We classify current existing surveys into five categories, including surveys on the energy efficiency of the whole cloud related systems, surveys on energy efficiency of the certain level or component of the cloud, surveys on all of energy efficient strategies, surveys on a certain energy efficiency techniques, and other energy efficiency related surveys. The survey and taxonomy on the energy efficient relevant surveys conducted in this paper provide comprehensive knowledge on the current level of this research filed. Furthermore, observations are exhibited by the statistic data graphs exacted from the investigated surveys. The observations hidden in those surveys on energy efficiency will make it clear the future research directions in the energy related research field. Future research directions are listed in the final section.

Classification and treatment of vehicle defect complaint data is an important link in the process of vehicle recall. Traditionally, the complaint data is classified by keyword matching method based on defect label library during the process of dealing with vehicle complaint data, which heavily relies heavily on the quality of the vehicle defect label library. The speed of traditional classification methods is rapid, but the accuracy is low. We transform the classification task of vehicle complaint data into a multi-label classification problem. Multi-label classification of vehicle defect information collection based on seq2seq model named VDIF-M is proposed in this paper. Firstly, a synonymous vehicle defect description label library is constructed based on the vehicle defect description data and vehicle domain corpus collected from various channels. Then a seq2seq model is proposed to solve the problem of multi-label classification of vehicle complaint data, which fuses the distribution relationship between labels. Substantial experimental results show that the proposed method outperforms previous methods in multi-label classification of vehicle complaint data.

Patent terminology relation extraction is of great significance to the construction of patent Knowledge graph. In order to solve the problem of long-distance dependency in traditional depth learning, a new method of patent terminology relation extraction is proposed, which combines attention mechanism and bi-directional LSTM model and with keyword strategy. Category keyword features in each sentence obtained by the improved TextRank with the patent text information vectorization added. BiLSTM neural work and attention mechanism are employed to extract the temporal information and sentence-level global feature information. Moreover, pooling layer is added to obtain the local features of the text. Finally, we fuse the global features and local features, and output the final classification results through the softmax classifier. The addition of category keywords improves the distinction of categories. Substantial experimental results demonstrate that the proposed model outperform the state-of-art neural model in patent terminology relation extraction.

Modern cloud data centers are virtualized for resource multiplexing and services consolidations. Virtual machines (VMs) residing in the same server cluster share the same hardware resources and power supply while they may have different QoS requirements for their services and applications. Moreover, the power consumption of the server cluster is highly dynamic since different virtual machines have different workloads due to different services requests. Cluster level power and QoS coordination is crucial for data center level energy efficiency coordination as well as high quality service provisioning. In this paper we propose the power and QoS aware multi-level resource coordination and scheduling in virtualized servers, i.e., the cluster-level power control layer, the VMs resource allocation layer, and the QoS optimization layer. This three-level controlling framework schedules and allocates hardware resource for QoS guarantee and cluster level power management. We use dynamic frequency scaling for QoS mitigating when power budget changes. The experiment results show that the proposed multi-level coordinated control architecture consumes 5.36% and 6.96% less power for web servers and computing intensive virtual machines, respectively while it can guarantee the response time of web server and execution time of computing tasks no more than those without the proposed controlling approach.

Understanding the characteristics of cloud workloads is the key to making optimal configuration decisions and improving the system throughput. However workload characterization of cloud, especially in a large-scale production environment, has not been well studied yet. To gain insights on cloud workloads, we collected a one-week workload trace from a 100-node cloud cluster which hosts 1082 virtual machines. We characterized the workload at the granularity of virtual machines and physical nodes, respectively. We concluded with a set of meaningful observations. The results of workload characterization are representative and generally consistent with cloud cluster for public IaaS service providers, which can help other researchers and engineers understand the performance and VM characteristics of the cloud in their production environments.

The energy problem is one of the serious problems in the current large-scale storage systems need to be addressed urgently. In order to reduce the energy consumption of cloud storage system, and to meet the performance needs of users, this paper purposed a cloud storage system integrated high availability green gear-shifting mechanism (HGLG): The frame designed a new data partitioning strategy, data replication management strategy and proposed energy gear-shifting mechanism based on the data partition and data replication management. Data partitioning strategy divide data into cold data, hot data, seasonal data and the new data and place it in the appropriate zone through green data classification strategy based on anticipation (AGDC). Depending on the nature of the data, the paper accordingly designed replica placement and replica number. Based on the above data partitioning, data replication management, this paper presents energy gear-shifting mechanism, which automatically gear-shifting through neural network model predicting follow-up period assignments. Experiments based on Grid Sim show that: the energy consumption of gear-shifting mechanisms cost effective, which saved about 43% average energy at the expense of about 1.6ms average response time and the maximum energy savings is about 70%.

Block level storage system has been widely used as it can provide users raw block device. On the basis of Orthrus system, this paper designs and implements a multiple volume servers block level storage system based on load-balanced strategy-Orthrus Plus. This system uses the dynamic load-balanced strategy based on genetic algorithm strategy to load balance the volume servers clusters and proposes an iSCSI pause and resume method to solve the breakpoint retransmission problem during load-balancing. At last, we design a hierarchical management mechanism based on volume life cycle at storage. Experiments show that the load balancing capability in Orthrus Plus is more accurate and efficient under the situation of larger amount of data, and the hierarchical storage strategy also increases the I/O throughout capacity.

A first-principles study has been performed to investigate the structural and electronic properties of theGaAs1-xBixsystem. The simulations are based upon the generalized gradient approximation (GGA) within the framework of density functional theory (DFT). Calculations are performed to different Bi concentrations. The lattice constant ofGaAs1-xBixincreases with Bi concentration while the alloy remains in the zinc-blende structure. The band gap ofGaAs1-xBixclearly shrinks with the Bi concentration. The optical transition of Bi dopant in GaAs exhibits a red shift. Besides, other important optical constants, such as the dielectric function, reflectivity, refractive index, and loss function also change significantly.

In order to solve the urgent issue of the energy consumption in the cloud storage system. An Energy-effective adaptive replication strategy (E 2 ARS) is proposed in this paper, in which data partition mechanism, minimal replicas determining model, replicas placement strategies and the adaptive gear-shifting mechanism are elaborately designed. We try to conserve the energy consumption while satisfying the users' desired response time by our E 2 ARS scheme. Mathematical analysis show that our E 2 ARS scheme will save energy consumption definitely when the system's workload is light or desir ed response time is loose. And the simulation experiment results demonstrate that through our E 2 ARS scheme energy consumption can be saved while with Qos satisfied and data availability