Abstract:With the development of the Internet, users tend to refer to online reviews before shopping, travelling, and dining. After that, they write reviews to express their own opinions. Online reviews are increasingly of great value. The significant guiding role of reviews playing in consumers' decisions has given rise to false comments, which we call review spam. The review spam refers to the comments written by users that do not meet the true characteristics of products, due to factors such as commercial profits and personal bias. Spammers imitate the writing style of true reviewers so that customers can hardly discriminate the review spam. Scholars at home and abroad use natural language processing techniques to detect review spam. From the perspective of feature engineering, review spam detection methods are divided into three types:the linguistic and behavior based, the graph based, and the representation learning based. This survey mainly describes the general process of review spam detection, summarizes feature designing of the models, and makes a comparison among three types of methods. Furthermore, the most commonly used datasets are introduced. Finally, it explores the research directions in the future.

Abstract:Coronary atherosclerosis is the most common disease of cardiovascular diseases in global. The mortality rate of human caused by the coronary atherosclerosis is gradually rising year by year. The main treatment adopted by the global medical institutions to reduce the pain of patients is the vascular stent implantation. Currently, in vivo angiography based on OCT performing a high resolution is gradually used in the examination and treatment of patients with cardiovascular disease. Hundreds of or thousands of IVOCT images of patients are produced out during each treatment or examination time. The method of traditional manual detection and marking for OCT images is inefficient and time-consuming. In response to such issues, researchers worldwide have done a lot of research and proposed many semi-automatic and automatic detection methods of the internal tissues and structures of vessels. The purpose of this paper is to comprehensively and systemically introduce the progress of vascular internal tissue detection researches based on IVOCT image and explains its principles.

Abstract:This paper describes the development status of immersive virtual reality (immersive VR) in the four periods, and analyzes its advantages and disadvantages. Furthermore, its development trend is discussed. In the future, the head-mounted display will be light weighted and comfortable, and the human-computer interaction will be more natural and convenient, and the immersive VR will be applied to lots of areas more generally.

Abstract:City gas load forecasting is significant to the operation of city gas networks. In consideration of the periodicity and nonlinearity of gas load data and the shortcomings of a single model, a hybrid model of Echo State Network (ESN) and improved RBF Neural Network (RBFNN) is put forward. First of all, kernel Fisher linear discriminant is utilized for dimension reduction. Secondly, we adopt ESN to do a preliminary prediction. Then, differential evolution integrated with gradient descent by encoding is used to learn and optimize the structure and parameters of RBFNN. Last but not least, the produced result of ESN is the input of RBFNN. It is validated that the proposed model has a higher precision and convergence rate compared with the initial combinational model.

Abstract:For the multi-attribute decision making problem with attribute weights being unknown and attribute values being hesitant fuzzy sets, this study proposes a multi-attribute decision making method based on prospect theory and rough set, which fully considers the influence of decision makers' psychological risk factors on decision results. Firstly, the positive and negative ideal points are used as reference points to calculate the prospect value function under each attribute and to define a new comprehensive prospect value, and a discernibility matrix is obtained according to the given threshold. Then, according to the discernibility matrix, attribute reduction is performed to determine the attribute weight. Finally, the weighted comprehensive prospect value of each alternative is calculated, and the TOPSIS method is used to rank all the alternatives. An example is given to illustrate the feasibility and effectiveness of the proposed method.

Abstract:The combinations of the Internet of Things and robots to make robots better serve the Internet of Things have been studied worldwide. This study puts forward an intelligent robot system based on ROS and Internet of Things. The hardware adopts stm32 sensor node, raspberry pi, OpenWRT router, Rplidar radar, c270 logitech camera, and so on. The software uses ROS operating system, Contiki, Tensorflow framework, Camshift algorithm, SLAM algorithm, and so on. Then we design the data acquisition of the perception layer, automatic traceability based on SLAM, voice control, object tracking of robot, object recognition, video monitoring, reverse control, and real-time display of data about perception layer on the web. After that, the relevant test environment was set up to test the related functions of the system and verify the feasibility of the system.

Abstract:The statistics of traditional and typical bus passengers have some shortcomings in accuracy and speed, and the effect of extracting target features is poor. This study proposes a bus counting system based on deep convolutional neural network to solve the crowd counting problem. The first thing to make a dataset is that all the datasets used for training are hand-labeled. And the bus camera angle is wider than the previous literature. This study first compares the effects of various deep convolutional neural network models on the whole body detection of passengers. Considering the detection rate and accuracy, the single-detector deep convolutional neural network model is used to detect passengers' heads. The simple online and real-time target tracking algorithm implements multi-target tracking of human heads, and the cross-region crowd counting method is used to count the number of passenger getting off the bus. The system accuracy rate reaches 78.38% and the operating rate is approximately 19.79 frames per second. the passenger count is achieved.

Abstract:Massive Multiple-Input Multiple-Output (MIMO), with giant array size and multi-dimensional array structure, has been widely considered as a key physical layer technique in future wireless communications. With regarding the large number of antenna elements, some new challenges and issues are arising. To solve these problems, a statistical channel model based on the spherical wave-front theory is proposed. Furthermore, the map-based ray-tracing algorithm with low complexity is used to compute the parameters of the proposed channel modeling. Finally, several statistical characteristics, such as delay spread and spatial distance, are given. The analysis results show that the proposed channel model is able to describe the main characteristics of massive MIMO channel.

Abstract:It is a research hotspot and difficult to display the Internet resources of ultra-high resolution effectively and quickly in the fields of satellite remote sensing, urban environmental monitoring and early warning, and smart city management center. This study proposes a parallel ultra-high resolution display technology that based on cluster which combines Web technology and cluster parallel processing technology perfectly. The technology overcomes many problems of the traditional centralized information processing display and the display resolution limit of DLP projection splicing technology, system splicing difficulties, slow information display processing speed, unable to different shared, etc., enables the real-time high resolution display of the Web resources. Through contrast experiments, the clustered Web parallel ultra-high resolution display technology proposed in this study can display Web resources efficiently and in high definition, which provides an effective solution for real-time and ultra-high resolution display of massive Web information content.

Abstract:For the problem that current methods unable to capture and analyze the system call parameters and return values, a system for real-time monitoring of system calls in the guest was established based on Nitro. The system capture and analyze fast system call entry and exit instructions by modifying hardware specifications and rewriting instructions. After capturing the system call entry instruction, the parameters are parsed according to the context information of the VCPU and the semantic template of the system call; after the system call exit instruction is captured, the return value is parsed according to the VCPU register information. Compared with the similar capture system call method, experiments show that the system can capture the system call sequence in the guest in real time, and obtain complete system call information including system call name, system call number, parameters, and return value. The system can also distinguish between system calls generated by different processes and brings no more than 15% performance overhead to the host.

Abstract:To improve the prediction accuracy of atmospheric pollutant SO₂, a combined forecasting model with optimal weights for atmospheric pollutant SO₂ was constructed, which was based on multiple air quality prediction (WRF-CHEM, CMAQ, CAMx) modes according to the principle of minimum square sum of combined forecasting errors of each single air quality prediction model in the past years. The actual observation data and the aforementioned three air quality modes prediction data of Chuxiong, Zhaotong, and Mengzi stations in Yunnan Province from January to May in 2018 were selected as the experimental samples. Then the multivariate linear regression method and the dynamic weight updating method were used to compare the prediction results with the optimal weighted combination prediction method proposed in this study under the same experimental conditions. The experimental results show that the predicted values of the proposed method are closer to the observed values than those of the other two methods, and the two error evaluation indexes are the smallest. Generally speaking, the combined forecasting model with optimal weights synthesizes the advantages of each single air quality forecasting model and improves the forecasting accuracy of SO₂.

Abstract:In view of the current railway industry's demand for fuel consumption control of diesel locomotives and improvement of fuel consumption management level, a locomotive oil quantity monitoring system based on NB-IoT is designed and implemented. The system uses high-precision ultrasonic liquid level sensor to obtain the liquid level of the oil, uses GPS to obtain the latitude and longitude in real time, uploads it to the server through the NB-IoT communication module, and stores it in the database after filtering algorithm, and thus monitors the real-time by oil quantity data, view historical oil quantity and fueling record using the browser access management system, realizes remote management of equipment. The test results show that the design can monitor the oil quantity, the precision is high, and the control of fuel consumption is improved. It has certain reference significance for the research of NB-IoT application and monitoring system.

Abstract:The face image data that can be obtained in the field of public security has grown rapidly. The traditional manual method to identify people has large workload, poor real-time performance, and low accuracy. This study designs a large-scale real-time face retrieval system. The system implements the real-time storage and retrieval of captured face images through the distributed platform Storm, and implements the storage and maintenance of large-scale unstructured face data through the distributed storage system HBase. The results of multiple experiments show that the system has a good speedup, good scalability, and real-time performance in the application scenarios of large-scale face image data retrieval.

Abstract:Aiming to solve the problem that the training teaching equipment of training course of distributed photovoltaic power station cannot meet the needs of practical teaching, a virtual simulation training platform for design, construction, and operation of photovoltaic power station is developed by using Unity3D engine with a real roof photovoltaic power station project as a typical case from the perspective of virtual simulation teaching. It also introduces the platform's functions, architecture, and key technologies involved in the development process. The application results show that the platform realizes the virtual visualization of the preliminary field exploration of the photovoltaic power station, design, construction, and operation and maintenance of the photovoltaic power station, and has a sound application value for the training of photovoltaic students and related technicians.

Abstract:At present, the research on the risk identification of power outage complaints and the customer sensitivity analysis in power grid companies is at its early stage. In order to effectively analyze the sensitivity of power outage customers, a sensitive customer classification algorithm based on the improved random forest algorithm is proposed. First, the data is preprocessed by methods of data cleaning, feature selection, and so on. Second, the SMOTE algorithm is used to increase the number of sensitive customers to solve the problem of data imbalance. Third, the representative feature space is selected by proportional random sampling. The Fisher ratio is used as the characteristic importance measure. Then, the random forest algorithm is used to recognize the customers that are sensitive to power outage. Finally, the experiments on real power outage data show that the proposed method not only has better accuracy and time performance but also can effectively deal with high-dimensional data with redundant features.

Abstract:A novel neural network for object recognition, CapsNet, uses dynamic routing and capsules to recognize novel state of a known object, while the input layer of CapsNet decoder increases when the number of categories increases, which means a relatively limited scalability. To overcome this weakness, we propose the Multi-branches Auto-Encoder (MAE) which gives coding vectors of every class to the decoder respectively letting the scale of decoder independent from the number of categories enhancing the representation capability of the proposed model. The experiment on MNIST shows that MAE is competitive in recognition and more powerful in reconstruction which means a more complete capability on representation.

Abstract:DAG task scheduling is the current hot topic. In task model of DAG, the order of task scheduling affect the service satisfaction of users on one hand, and also affect utilization rate of cloud service resources on the other hand. High efficient task scheduling algorithm may strengthen the resources distribution of the multi-core and the parallel computing ability. HEFT algorithm and CPOP algorithm are of lower efficiency in related task scheduling. Based on HEFT algorithm and CPOP algorithm, a dependent task scheduling model and task scheduling algorithm IHEFT (Improvement Heterogeneous Earliest Finish Time) algorithm are proposed in this study. The IHEFT algorithm mainly optimizes two aspects:task ordering and task scheduling. The variance of task scheduling cost on every processor core and the average communication overhead are the basis of task ordering. In the stage of task scheduling, task duplication of some nodes in DAG with some conditions can make full use of heterogeneous processor resources and shorten the completion time of task set. Experiment results show that the IHEFT algorithm performs more performance than the HEFT algorithm and the CPOP algorithm in terms of the task scheduling Makespan, the average waiting time and the average value of Slack.

Abstract:In recent years, with the improvement of the pace of life and the rapid development of the Internet, people are more inclined to communicate with the short text on many social platforms, and then some people can disturb the network's green environment by releasing the spam texts to hinder the normal social intercourse. In order to solve this problem, we propose a method of spam text detection based on optimized BP neural network and social platform. Through this method, the spam text filtering on the social platform is realized. First of all, through the stuttering participle and to stop word to construct keyword data set. Secondly, the keyword vector of the text expression is used to compute the weights of each keyword so as to reduce the dimension of the text vector and obtain the eigenvector. Finally, based on this, the BP neural network classifier is used to classify the short texts, and the spam text is detected and filtered. The experimental results show that with this method, the average classification accuracy for the 1000 dimensional text feature vector reaches 97.720%.

Abstract:In order to meet the requirements of Java static distributed detection system for decoupling and subcontracting Java program source code package, Java source code file dependency analysis method was proposed. This method is part of single-task multi-node parallel-running distributed detection system which could solve the problem that it takes long time for single-task single-node single-process code detection. The method extracted the Java source code file text information by generating its abstract syntax tree. Then, the method traverses and parses the abstract syntax tree to obtain the file's dependent classes which were not declared in the file. Finally, the dependency relationship between two files was obtained by locating the files where the dependent classes were declared in. The directed cyclic graph with no incident edge vertices was proposed to represent the dependency graph among files. The Java program source code package was decoupled by analyzing the dependency graph. At last, the feasibility of the proposed Java source code file dependency analysis method is verified by two experiments. The first experiment is the step-by-step result analysis of a sample program package. The second experiment is verifying the correctness of decoupling result of several open source tools' source code packages.

Abstract:In view of the fact that the current algorithm can not meet the needs of fabric defect classification detection with periodic pattern characteristics, a deep convolutional neural network fabric defect detection algorithm based on Fisher criterion is proposed. First, a small Deep Convolutional Neural Network (DCNN) is designed by using depthwise separable convolution. Further, the Softmax loss function of DCNN adds Fisher criterion constraint and updates the whole network parameters through gradient algorithm to get Deep Convolutional Neural Network (FDCNN). Finally, the classification rates of TILDA and pink plaid fabric database were 98.14% and 98.55%. The experimental results show that the FDCNN model can not only effectively reduce network parameters and running time, but also improve fabric defect classification rate.

Abstract:The development of speech recognition is changing with each passing day. At the same time, the existing research results show that there is more complementary information in acoustic characteristics. In this study, a trajectory based spatio temporal spectral speech emotion recognition method is proposed. Its core idea is to get spatial and temporal descriptors from the speech spectrum, classify and identify dimensional emotion. The experiment using the exhaustive feature extraction shows that the proposed method is more robust in the noise condition than the MFCCs and the fundamental frequency extraction methods. In the 4 classes of emotion recognition experiments, the comparison of non weighted average feedback is obtained, and more accurate results are obtained. And, the voice activation detection is also improved significantly.

Abstract:For the multi-attribute group decision-making problems, where the information of the attribute weights and the expert weights is completely unknown and the preference information is in the form of hesitant 2-tuple linguistic, a multi-attribute group decision-making method based on the prospect theory and the grey relation analysis is proposed. Firstly, the weights of the experts are determined by the matrix vec operator and the grey relation analysis, and the weights of attributes are calculated by the maximizing deviation method. Subsequently, a comparison method of the hesitant 2-tuple linguistic elements is given, and the positive and negative ideal solutions based on that are determined and used as the decision reference point. Then the hesitant 2-tuple linguistic prospect value function according to the prospect theory and the grey relational coefficient is acquired, and then the ratio of the gains to losses of the alternatives is obtained, and the alternatives are ranked accordingly. Finally, the proposed method is applied to a numerical example of investment decision, and the results show the rationality and effectiveness of the method.

Abstract:In this work, the migration timing of virtual machines is studied for frequent migration of virtual machine in cloud data centers, an adaptive migration trigger method of virtual machine based on improved exponential smoothing prediction is proposed. A combination of dual threshold and prediction is applied to the strategy. First, the load prediction is triggered by continuously determining the load state. Then, the host load state at the next moment is adaptively predicted based on the historical load value, and finally the virtual machine migration is triggered. This method not only achieves host load balancing, but also improves migration efficiency and reduces energy consumption. Experiments show that the method reduces the energy consumption and the number of migration by about 7.34% and 58.55% respectively, which has sound optimization effect.

Abstract:The short life of battery becomes the bottleneck that greatly undermines user's experience of smart phones due to the high power loss on display panel. To overcome the challenge, a low-power and high-quality display driving algorithm based on two-line model is proposed in this study. Firstly, the original image is converted from the RGB space to the YUV space. Subsequently, the average and maximum luminance values are calculated to enable the two-line model to be established based on the corresponding average and maximal luminance values. Finally, the YUV space is converted back to the RGB space and the renewed image is obtained after generating the renewed luminance under the assistance of the two-line model. The experimental results show that the MSE of image processed by the proposed algorithm is 30.9% and 29.9% lower than that processed by the NPC and the ACSC, respectively, with the lowest power consumption. Moreover, the proposed algorithm has been successfully verified on an FPGA board without causing a significant degradation of visual effect.

Abstract:Aiming at solving problems of data redundancy and low query efficiency in the storage of mass social work data, this study proposed an effective partition-based neighbor sorting algorithm. The social data collected by different channels and stored in different storage methods were integrated to form a massive data set that can be stored in a two-dimensional form. The partitioning idea was used to segment the massive data set to clusters; the improved neighbor sorting algorithm was used for each cluster to obtain the final similar duplicate record detection results. The experimental and comparative analysis results show that the combination of partitioning and neighbor sorting algorithm not only improves the time efficiency of similar duplicate records detection of massive data, but also improves the detection accuracy.

Abstract:With the update of intelligent equipment and the improvement of data storage capacity, manufacturing companies have achieved a large amount of pipeline data in the manufacturing process of their products. How to utilize these data has always been a difficult problem in the industry. Depending on the actual production data of manufacturing enterprises, this study establishes a product failure identification model based on FTRL (with Logistic Regression) and XGBoost algorithms through detailed exploratory data analysis, then uses cross-validation methods to optimize it according to MCC metric which is suitable for unbalanced datasets. The experimental results show that the model has a high efficiency and high accuracy of fault prediction for large-scale (not only large sample size but also large feature quantity) unbalanced production pipeline datasets. Based on this model, we can build a smarter product fault detection system, which effectively reduces the operating costs of the enterprise and also spurs profit growth.

Abstract:The networked software is composed of heterogeneous Web services which are distributed on the Internet. Service trustworthiness is in constant change under the influence of dynamic and open environment. To solve the problem that Web service trustworthy evaluation is difficult to adapt dynamic environment, this study developed a dynamic trustworthy evaluation based on information entropy and correction metrics. It mined deeply the objective information in fuzzy matrix and modified the subjective weights that are susceptible to human factors. Considering the impact of the running environment on the trustworthiness of the current service, it added correction metrics and external metrics affected by feedback to improve self-adaptability of the evaluation model. Finally, simulation experiments based on a map service show that the DTMIECM model possesses feasibility in trustworthy measurement. It can adapt to unstable environment and reconfigure service to improve trustworthiness of the whole system.

Abstract:The tendency analysis of the judgment results in the judgment documents is the premise of completing the lawyer recommendation system. How to effectively realize the tendency analysis of judgment results has been focused on. This study proposes a biased analysis model based on attention mechanism and BiGRU. First, training the word embedding to obtain a word embedding table. Then, by looking up the word embedding table, the document data is transformed into a word embedding sequence, and the word embedding sequence is used as input to train the judgment result orientation analysis model. The experimental results show that the attention mechanism and BiGRU algorithm have certain effectiveness in the analysis of decision results. The model can make a reasonable judgment on the propensity of the judgment results in the judgment documents, and provide a reasonable scoring basis for the realization of the later lawyer recommendation system.

Abstract:Most of the research on the coverage of existing equipment area aims at two-dimensional undirected sensing area. The location of equipment is often set randomly. The coverage of complex space environment is low, which is difficult to meet the needs of practical applications such as safety monitoring. The study puts forward the algorithm of adjusting the direction of the equipment in combination with the virtual force and the regional weight in the three-dimensional space region, increasing the effective coverage of the area, and finally giving the deployment of the additional equipment based on the greedy strategy to further enhance regional coverage. The experimental results show that the proposed scheme can greatly improve the regional coverage at a small cost.

Abstract:The Information Centric Networking is a brand new network architecture that addresses and routes content names. However, there is no efficient and concise solution to realize the movement between producers and consumers. This study proposes a mobility support scheme based on the POF-ICN architecture. It registers location information, mobility management, session management, and controller collaboration to maintain related business information and plan paths in order to support the movement between producers and consumers. The system mainly uses the device to register the location information on the global resolution system, and then the mobile management, session management and the cooperation between the controllers. This work is to maintain related business information and planning paths, in order to support mobility. Through related experiments, we found that almost all requests successfully complete the response after the consumer successfully sends the request. And the pre-planned path can effectively improve the request response rate when the network is relatively congested. This research proves that the proposed mobility support scheme based on the POF-ICN architecture can implement the mobility support between producers and consumers concisely and efficiently. Moreover, the mobile packet loss rate can be effectively reduced by the pre-planned path.

Abstract:In this study, a novel identification method of the pointer instrument is proposed, which is implemented by using two-dimensional code matching information. Firstly, a high quality instrument state image and a two dimensional code image are acquired together. And the two-dimensional code location points and the type of instrument are obtained by matching the two dimensional code. Secondly, the instrument image is quickly corrected by the two-dimensional code location information. Then, using the geometric position relationship between the two-dimensional code and the instrument, the region of the dial plate in the instrument image is extracted quickly. Finally, according to the information of the instrument type, the corresponding instrument identification algorithm is selected to recognize the instrument readings quickly and accurately. The experimental results show that the proposed method can effectively improve the accuracy of the recognition of the pointer meter reading, especially for the instrument images of the complex background, which can provide an effective way for the pointer instrument identification in the power system.

Abstract:Recording and analyzing the data generated by online learners on the Internet and providing accurate and personalized services is an important aspect of online education. This study takes the daily learning data generated by learners on the teaching platform as a sample, synthesizes its five most representative influencing factors, classifies samples by Learning Vector Quantization (LVQ) neural network, and obtains online learning academic performance prediction data based on BP network. The genetic algorithm is used in the model to effectively optimize the weights and thresholds of the BP network, which accelerates the convergence of the model while improving the prediction accuracy. Finally, compared with the other two models, the results show that the model's prediction results are basically consistent with the real performance distribution. It has a high degree of credibility and provide a decision-making basis for effective prediction of learning status, which has certain value in engineering application.

Abstract:Considering the growing development of e-commerce platforms, the use of artificial classification of clothing classification cannot meet the current needs. In this study, based on the actual application scenarios, the design is improved in three aspects:the interference of background factors, the key position information of the garment image, and the hardware requirements of the algorithm model operation when classifying the garment image. Accordingly, it is proposed that to remove background interference, to use of local information of images, and to lightweight processing of the model. Finally, on the premise of satisfying the accuracy, the algorithm model that can be operated in the ordinary low-configuration PC terminal is obtained, which improves the work efficiency and saves the cost.

Abstract:In order to achieve fast and accurate image retrieval for large-scale clothing image sets and break through the limitations of current conventional retrieval methods, this study proposes a new deep learning model:Fashion-16 clothing image retrieval model. Based on the idea of first classification and intra-class retrieval, based on the powerful image feature extraction ability of VGG-16 model, the convolutional neural network Softmax classifier is used for classification, and the nearest neighbor search is performed for the idea of locally sensitive hashing under the same category. An image retrieval model correction for clothing category attributes is implemented. The experimental results show that the model has good stability, accuracy, and retrieval speed, and has practical value and research significance.

Abstract:This design uses STC89C52RC and STM8S005 as the core control, contains the digital tube display module, ultrasonic module, button module, infrared ranging module, proximity induction, mobile phone charging, Bluetooth connection to play music, putter and voice recording device and app console lamp. This design adopts the humanized way, can rise the height range of 50 cm to 110 cm, achieve the human body's standing and sitting, desk can real-time response to the human body's posture, real-time measure height, and design APP to achieve additional music playing, LED lighting and colorful lamp changes, and other functions.

Abstract:This study investigates event-triggered H_∞ filtering problem for discrete-time multi-agent systems with switching topologies. A novel distributed event-triggered control scheme is constructed to determine whether each agent should transmit the current sampled data to the filter, thus effectively save network resources. Considering the existence of network-induced delay and modeling the switching of network topologies by a Markov process, the distributed filtering error system is modeled as a closed-loop system with multiple time-varying delays by using the event-triggered control scheme and the proposed H_∞ filter. By employing Lyapunov-Krasovskii functional and linear matrix inequality method with multi-interval upper and lower bounds information, some sufficient conditions and the design method of H_∞ filter parameters are obtained to guarantee the closed-loop system to achieve asymptotic stability with an H_∞ performance index. Finally, two numerical examples are given to illustrate the effectiveness of the proposed method.

Abstract:In order to solve the problem of slow convergence of traditional BP (Back Propagation) neural network, through the BP neural network build the fire point prediction model, we use an adaptive learning rate method to improve the BP neural network, by comparison, the algorithm converges faster, and the output of the model achieves the desired effect. At the same time, an improved algorithm is realized by using the dynamic reconfigurable technology of FPGA. Through the simulation and results test, the design greatly reduces the prediction time on the basis of the prediction results and provides a theoretical basis for environmental prediction and detection trajectory planning.

Abstract:When the traditional topic model method is used to analyze the sentiment of the topic model for short text corpora such as comments in the medical service platform, the problem of poor context dependency may occur. A WLDA algorithm based on word embedding is proposed. The word w* trained in the Skip-Gram model replaces the word w` in the Gibbs sampling algorithm in the traditional LDA model, and the parameter λ is introduced to control the resampling probability of the words during Gibbs sampling. The experimental results show that the subject model has a high degree of consistency compared to similar topic models.