Abstract
China’s information technology development is rapid; the organic combination of science and technology and education has promoted the reform of the education system; the education platform supplemented by mobile devices provides students with a new learning mode; student learning is no longer limited by the traditional education system; and learners can use mobile terminals for online learning, self-control learning steps, arrange learning time, and be able to combine their own deficiencies to strengthen learning content. Therefore, this paper discusses the design of English education teaching system in the Internet of Things technology environment, in order to provide students with a good learning environment and improve students’ English integration strength. This paper analyzes the research status of mobile learning at home and abroad, expounds the construction principles of mobile English learning system, and designs and studies the overall structure and functional modules of the system. The application of the system can replace the traditional English learning method and can enable students to improve their English ability through mobile devices. In view of the problems of poor teaching effect and high energy consumption in the traditional Internet of Things education platform, this paper puts forward a teaching effect evaluation method of Internet of Things education platform based on long-term memory network. By analyzing the current situation of the Internet of Things education platform, including its development and structure, this paper constructs the evaluation model of the combination of long-term memory neural network model and gray model and realizes the evaluation of the teaching effect of the Internet of Things education platform. Finally, through the study of the model, the teaching effect of the Internet of Things education platform is evaluated. The experimental results show that when the method is used to evaluate the teaching effect, the operating energy consumption accounts for 84% of the total energy consumption of the system, which proves the effectiveness of the method.
1. Introduction
The Internet of Things is closely related to the Internet, mobile communication networks, sensor networks, etc. and is an information carrier based on the Internet and traditional telecommunications networks, which can play a role in interconnection and interoperability and then find objects through information indexing [1]. From a technical point of view, the Internet of Things project has 3 parts: (1) external perception; (2) combination of nodes and wireless networks in system design to transmit and perceive information; (3) information processing and feedback control. From the perspective of combining technical levels, the intelligent development of the Internet of Things is the core of technology. Through intelligent induction, recognition, perception, etc., data can be effectively processed. It can be said that the Internet of Things technology is both “intelligent science and technology” and “computer science and technology” embodiment. For this reason, China’s major universities will be those of “intelligent science and technology.” This discipline is linked together to conduct deep-level research [2].
Since the 1970s, there has been an increased emphasis on intelligent teaching systems. In particular, the United States and other Western countries attach great importance to the development of intelligent education systems. These countries have strong economic strength, science, and technology to not only drive the development of intelligent market industries, but also optimize the design of intelligent education systems [3]. Combined with the development of the situation at that time, the intelligent teaching system in the United States is mainly for physics, chemistry, and other science services. Intelligent science services can improve professional teaching services through the system simulation data. With the development of the Internet of Things technology, intelligent teaching system is gradually applied to school teaching, with remarkable achievements [4].
In contrast, China’s intelligent teaching system still needs to make a breakthrough, further strengthen and improve the service function of the intelligent teaching system, and provide effective services for actual teaching. At present, China’s intelligent teaching system has made a breakthrough in technology, mainly reflected in artificial intelligence technology assisting teachers in their daily work [5]. In May 2019, the first International Conference on Artificial Intelligence and Education aimed to explore the development strategy of intelligent education and to promote the development of intelligent teaching and scientific planning for it, thereby accelerating educational innovation [6]. Combined with traditional education, intelligent education is a new type of education system based on computer technology, which emphasizes the organic combination of intelligent technology and education, teachers, teaching, etc., which can improve the education system, thereby strengthening the quality of teaching and improving students’ interest in learning, and can carry out systematic evaluation, so that students can learn independently, understand concepts, and master skills. Especially in English education, it is necessary to innovate and strengthen students’ English ability [7]. At present, the application of computer-aided tools in English-assisted teaching is mainly reflected in multimedia and teaching application software. With the application of computer multimedia technology, English listening teaching has been more abundant in teaching form and content and more able to attract students’ attention. Through the use of multimedia to play the listening material, this set of graphics, text, and sound in one form can stimulate students’ interest in learning and increase students’ logical thinking and independent learning ability [8].
2. State of the Art
After the 90s of the last century, China began to study virtual simulation technology, but the real application of this technology in practical teaching is in this century, with the first field of application being higher education [9]. After more than 20 years of rapid development, virtual simulation technology has been widely used, especially in the national vocational college skills competition, teaching ability competition, and other information teaching competitions, thereby improving the overall level and promoting the cultivation of students’ innovation and entrepreneurship capabilities. In order to analyze the application status of virtual simulation technology in China in teaching, some literature was collected from electronic journal websites such as CNKI and Wanfang database [10]. The number of applications of virtual simulation technology in disciplines has shown a significant increase trend every year, and most of them are disciplines with obvious practical characteristics or disciplines with obvious applications, such as medical fields, industrial control, mold manufacturing, and scientific experiments, as well as some programming and design fields, involving electronic information. Processing and manufacturing virtual simulation technology applications have gradually become more and more in the past three years [11]. On the application of virtual simulation technology in the Internet of Things profession, the following is summarized.
Wang Jing et al. conducted research and analysis on the construction and practice of the virtual simulation experimental teaching system of the Internet of Things engineering major of the school of medicine; funded the research and development of medical network system simulation software, that is, MNSS software; and integrated the characteristics of diversification and modularization into the actual teaching, seeking a way to cultivate innovative composite talents in this profession and also providing a new way for the cultivation of college students’ practical ability [12].
Tang Haitao et al. introduced the basic situation of the experimental platform construction of the Virtual Simulation Experiment Teaching Center of the Internet of Things of Jilin University and used the existing equipment and resources to build a relatively realistic teaching curriculum [13]. The teaching center contains 12 experimental courses that require the help of virtual simulation technology, such as RFID tag design and wireless sensor simulation experiments.
Gao Liqin et al. believe that traditional experimental teaching activities are easily limited, and there is a shortage of equipment resources, which makes it difficult to meet the needs of some new experimental teaching [14]. With the continuous development of information technology, virtual experimental teaching has also become a new development trend, helping to solve many problems in traditional experimental teaching activities, and the application of three different simulation software systems is introduced in this study.
Xing Dan et al. analyzed the existing problems in the RFID experimental course in the Internet of Things project, such as the relatively high cost and limited experimental conditions; proposed the use of open source virtual simulation platform to complete the relevant practical teaching activities; introduced the basic situation of the platform in the study; and believed that this virtual simulation platform can help students participate in engineering practice and find and solve related problems [15].
Yu Xiao et al. analyzed some common problems in the teaching of the Internet of Things in medical schools; took the Internet of Things engineering major of Xuzhou Medical University as an example; introduced the concept of “microteaching”; developed the medical network system simulation software; formed a fragmented, interactive communication learning mode; and laid the foundation for the improvement of the innovation ability of the students of the Internet of Things engineering major in the medical school and the cultivation of independent learning behavior habits [16].
Li Hongliang et al. discussed 12 existing virtual simulation experimental courses, including three major aspects; analyzed the use effect of the virtual simulation platform in experimental teaching; and found that this teaching method can greatly improve the efficiency of experiments and reduce the time spent by students in actual experiments [17].
Zhao et al. listed the application process of BP neural network model in the evaluation of female entrepreneurship education and established the BP neural network model based on BP neural network. The maximum relative error between the actual value and the expected value was about 1.64%, and the comprehensive evaluation value was 92 points, effectively avoiding the traditional model [18].
It is true that our country already has a variety of online English education websites. But many problems still seriously affect students’ use of the system. Different regions choose different teaching textbooks in the intelligent system, resulting in different textbooks and learning contents. Different regions use different versions of textbooks and training programs, resulting in the differentiation of educational ideas. At the same time, for middle and high school students, the high examination papers in different regions are also different, and the learning content lacks systematic construction, so a single platform can only achieve the transmission of specific knowledge and cannot be trained according to the student’s region.
In view of the problems of poor teaching effect and high energy consumption in the traditional Internet of Things education platform, this paper puts forward a teaching effect evaluation method of Internet of Things education platform based on long-term memory network. By analyzing the current situation of the Internet of Things education platform, including the development and structure of the Internet of Things education platform, this paper constructs a combination of long-term memory neural network model and gray model to evaluate the teaching effect of the Internet of Things education platform. The specific process is shown in Figure 1.
3. Methodology
3.1. Introduction to IoT Technology
International development and global economic integration development have promoted the innovation of Internet technology. With the support of Internet technology, the formation of an online education system strengthened the development of modern education. Although the technology continues to innovate and online education software emerges in an endless stream, due to the limitation of transmission rate, there are still problems in the application of teaching courseware and network speed, affecting online learning.
In addition to the sharing, virtuality, interactivity, space-time expansion, and other characteristics of the Internet, the Internet of Things also has some other characteristics, which extend and expand the impact of the Internet on our production and life [19].(1)The perception of underlying data is the basis of Internet of things technology. In the perception layer of the Internet of Things, the characteristics presented are large amounts of data, with a wide variety, so the configuration of different types of sensors is very effective. Different types of sensors perceive the information content, and information format is different; each has its own use. In addition, the sensor also has the characteristics of real time. The information of the perception layer changes quickly and is constantly updated, and the sensor must always capture the information. In such an environment, people can obtain the vast amount of data they need from the virtual and physical worlds [20]. Data is the foundation of information services, and it can provide strong support for human decision-making and judgment.(2)The important foundation and core of the Internet of Things technology is still the Internet. After the Internet of Things perception layer collects information, it must pass through the transmission layer for data and information transmission. Through the integration with Internet technology, the information collected by the perception layer is transmitted in real time and accurately. In the transmission process, the security and effectiveness of the information must be guaranteed, so the transmission layer should be like the Internet transmission layer. A perfect transmission mechanism needs to reach such a consensus agreement first to ensure the security and effectiveness of the transmission process [21]. With a ubiquitous network, people can get data anytime from anywhere, any person, and anything in real time. It also provides people with a channel to send the data or information that people need to release to any place, any person, and anything at any time, to achieve an interactive scene in which the upward collection and downstream control are integrated.(3)The development trend of Internet of Things technology is to achieve intelligence in the world, which is bound to require the support of various types of high-end equipment. Cloud computing platforms can solve this problem very well and achieve efficient data sharing and exchange. For the characteristics of many customer needs in the Internet of Things, it is necessary to propose specific ways and methods to solve specific problems, and the virtualization that cloud computing technology offers can achieve this goal [22]. Only through the support of high-end technology can the Internet of Things technology achieve cross-industry and intelligent information services for the whole world.
3.2. LSTM Neural Networks
Long short-term memory neural networks, often referred to as LSTM, is a special kind of recurrent neural network that learns long-term dependencies. After the improvement and promotion of many researchers, LSTM has been widely used. It is composed of Hochreiter and Schmidhub [23].
LSTM is widely used for many sequence tasks (including natural gas load forecasting, stock market forecasting, language modeling, and machine translation) and performs better than other sequence models such as RNNs, especially when there is a large amount of data. LSTMs are carefully designed to avoid gradient vanishing problems with RNNs. The main practical limitation of the vanishing gradient is that the model cannot learn long-term dependencies. However, by avoiding the disappearing gradient problem, LSTMs can store more memories (hundreds of time steps) than conventional RNNs. LSTM has more parameters than an RNN that maintains only a single hidden state, which gives it more control over which memories are saved or discarded at specific time steps. For example, the hidden state must be updated at each training step, so the RNN cannot determine which memories to save and which to discard.
LSTM remembering more historical information is actually its default behavior without deliberate learning. In addition to the external RNN loop, the LSTM also has an internal “LSTM cell” loop. Similar to ordinary recurrent neural networks, each cell has the same inputs and outputs, and there are more parameters and gated units that control the flow of information. Figure 2 shows the internal structure of the LSTM.
The output of the LSTM cell can also be controlled by an output gate and (using a sigmoid unit as a gating):where are bias, input weights, and cyclic weights of the forget gate, respectively.
3.3. GM (2,1) Gray Predictive Model
The gray prediction principle is to follow the established development law of the system, establish a general gray differential equation, determine the equation coefficients through the method of data fitting, and finally obtain a complete gray model equation for the prediction of the long-term development of the system [24]. The gray model is suitable for data fitting of short series, and long-term dynamic characteristics can be obtained through simple information. When the target settlement is monotonically increasing and bounded and presents an S-shaped curve, it can be considered to be in line with the saturated S-shaped curve, and the GM (2,1) prediction model should be used.
3.4. Principles of Fast Uncontrolled Multiobjective Genetic Algorithms
The common weighting criterion of model fusion is based on the criterion for determining the weight coefficients of the combined model with the smallest sum of squares of the prediction error, which is simple to calculate and widely used, but its disadvantage is that when the minimum root mean square error cannot be met with other conditions at the same time, the optimal scheme under multiple accuracy requirements cannot be obtained [25]. Therefore, it is necessary to use a multiobjective optimization method to determine the optimal weight distribution of the combinatorial model. For single-target problems, genetic algorithms show their superiority. But in the multiobjective optimization problem, the biggest problem that people first encounter is how to measure the fitness of an individual. For example, in a class, every student has three grades in addition to the number of words. If we give a certain weight to the number of words, then we can weight the grades of the three subjects outside the number of words to sort the grades of all students, so that the problem degenerates into a one-goal optimization problem.
The NSGA-II algorithm, a fast nondominant multiobjective genetic algorithm with an elite retention strategy, is also called a non-inferior hierarchical genetic algorithm based on Pareto optimal solution, which adopts a better accounting strategy than the underlying NSGA algorithm and reduces the overall time of the algorithm [26]. The algorithm uses density estimation operators and crowding degrees to judge the non-inferior level of each individual, so as to quickly perform non-inferior ranking, maintain population diversity, and reduce the computational complexity. The algorithm also introduces elite strategies to avoid the loss of the best individuals, which improves the algorithm’s computational speed and robustness.
3.5. Model Evaluation Indicators
There are limitations in a single accuracy standard, so the evaluation terms of model accuracy select three indexes: average absolute error (MAE), average relative error (MRE), and root mean square error (RMSE) for accuracy evaluation [27]. In order to reflect the integrity of settlement prediction, the posterior difference ratio C is added; the small error probability p characterizes the correlation degree between a single data item and the overall data, and the gray model increases the gray correlation degree to test the accuracy, which is calculated as follows:where 2–4 measure the deviation between the predicted value and the true value and 5–8 are used to test the accuracy of the gray model (Table 1).
4. Result Analysis and Discussion
4.1. LSTM Model Creation
The long short-term memory network (LSTM) is a special type of RNN network designed to solve the problem of long dependencies. The network was introduced by Hochreiter and Schmidhuber (1997) and was improved and popularized by many people. Their work has been widely used to solve a wide variety of problems until now.
When processing the settlement data, the LSTM model needs to process the long-term sedimentation data as a time series as a learning sample. This paper takes the sedimentation value of the first three time points as the eigenvalue value, takes the data of every four time nodes as a set of eigenvalues and output values, takes the first 70% of the sedimentation value of the entire data source as the training set, predicts the post-settlement data, and compares it with the measured sedimentation. For gray models, the same partitioning rules are followed, using only the data for the training set portion of the fit. For the same set of data, a long- and short-term memory neural network model and a gray prediction model are established to predict the sedimentation value of the future period of time. Among them, the long- and short-term memory neural network model adopts the Python language, the LSTM module in the Keras library based on TensorFlow, and the preprocessing module in Scikit-Learn for modeling. The gray forecast model is calculated using MATLAB.
LSTM neural networks need to use time series data with consistent time intervals, so the data needs to be interpolated in advance. In the time series, the last 30% is taken as the test set and the remaining data as the training set [28]. For the gray model, only the above training set data is taken for fitting, and the fitting results are brought in to calculate the subsequent predicted values. Before training, time series data needs to be normalized and regularized to eliminate data noise, avoid model overfitting, and increase model sensitivity.
The time series is processed in the LSTM model by defining the time step parameters, that is, using several sequence values before the target sequence as its eigenvalues. The LSTM model has a total of 7 parameters, where the input feature dimension is related to the time steps, and the initial time steps are set to 3, which means that a prediction is considered to be related to the previous three historical data items. The initial time step of the input and output layer is 1. The activation function uses the sigmoid function. The model has 100 iterations, 1 hidden layer, and 16 hidden nodes. Batch size is set to 1, which is random training. Setting the loss function to mean squared error, i.e., the squared loss, gives a heavier penalty for predictions that deviate more from the measured value. By analyzing the historical data of the loss of each operation cycle of the model, it can be determined whether the model is underfitting or overfitting, and the relevant parameters can be modified and adjusted. Gray predictions are solved using conventional methods.
4.2. Combined Model of Long- and Short-Term Memory Neural Network
In order to combine the advantages of long-term and short-term memory neural networks and gray models, it is necessary to construct a combined model of neural networks and gray models and determine the weight coefficients in the combined models based on fast nondominant genetic algorithms.
Using a genetic algorithm to calculate the combined model weights, the two fitness functions are defined as follows:and f1 and f2 represent the ratio of root mean square error to posterior difference of the predicted settlement value in the last training stage before prediction, and the optimization goal is to obtain a weight distribution scheme with the lowest ratio of root mean square error to posterior difference of the predicted value of the combined model. In this algorithm, population selection is performed using the bidding method, binary population crossover is simulated using real number coded crossover operation, and population variation is achieved using polynomial variation method.
The time series classification model of classroom teaching behavior based on LSTM is shown in Figure 3, including input layer, masking layer, LSTM layer, and classifier layer. The first layer of the model is the input layer, which is mainly responsible for inputting the training data into the LSTM network, in which the input data is the classroom teaching behavior coding data after data preprocessing. The second layer is the masking layer, and the function of the masking layer is to automatically “filter” the 0 values in the encoded data passed by the input layer. The third layer is the LSTM layer, which is mainly responsible for extracting the key discriminative information from the classroom teaching behavior data. At the same time, in order to prevent overfitting, this paper uses Dropout to reduce the complexity of the model. The fourth layer is the classifier, which is mainly responsible for calculating the class probability distribution vector of the classroom teaching behavior coding data. This paper uses the Softmax function to realize the classification operation and outputs the category probability distribution vector of the classroom teaching evaluation index.
4.3. Evaluation Model of the Information Sharing Method for English Language Teaching in the Internet of Things
4.3.1. Determining the Number of Neural Network Layers
In 1998, Robert Hecht-Nielsen verified that a three-layer neural network is a complete network that can achieve full reflection, provided that the verification is a continuous function in any closed interval, and then through the process of infinite clamping of an implicit layer BP network, the final result is from n-dimensional to m-dimensional reflection. Kolmogorov’s theorem reveals that even three-layer neural networks have a wide range of applications and strong performance, so a three-layer neural network can approximate a set function with arbitrary error, provided that the number of hidden layer nodes is not limited. Increasing the number of layers of the hidden layer has advantages and disadvantages. Therefore, to obtain a lower training error, the optimal choice is to create a three-layer neural network model.
4.3.2. Determining the Number of Neurons in Each Layer
The number of neurons in the input layer is an important feature of the neural network, on the basis of which the two are equivalently unified, and the evaluation index is 21 parameters for the sharing of English language teaching information for the Internet of Things (participating in provincial competitions and winning awards, participating in national competitions and winning awards, participating in scientific research projects, publishing journal papers, etc.). Therefore, the number of neurons in the input layer is n = 21.
The numbers of neurons in the hidden layer and output layer are determined as follows: The actual output value is unique according to the network, so the number of neurons in the output layer is also unique, that is, m = 1. Compared with the number of neurons in the implicit layer and the number of neurons in the input layer and output layer, they are susceptible to interference and conflict of the peripheral environment. If the number is small, the amount of information is small; if the number is large, it leads to poor fault tolerance, too long training time, and “excessive coincidence.” So there is an optimal value, derived from the empirical formula:
By Table 2, the number of neurons in the implicit layer is analyzed as 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, and 13. At this time, the training error decreases continuously. There is a correlation; the test error rises first and then falls when the value is 10, 11, and 12, and the overall number of neurons in the hidden layer being 11 is the best choice.
The learning rate directly affects the efficiency of the training and testing of neural networks. The learning rate is equivalent to . There are pros and cons; when the learning rate is larger, the weight is large and the convergence is fast, resulting in large network fluctuations. A small learning rate causes the network efficiency to be uncertain and the convergence to be slow. The introduction of momentum items can solve such problems.
Based on the evaluation results of this paper, the error and number of training steps in Table 3 were analyzed at a learning rate of 0.01.
Form Table 4, the six representative training functions are Levenberg–Marquardt, Rprop, scaled conjugate gradient, one step secant algorithm, gradient descent, and gradient descent with adaptive learning rate and momentum factor. The training results of different training functions are compared in terms of the number of training steps and performance, and the performance advantages of different training functions are highlighted.
Momentum factor plays an important role in the neural network. In particular, in the training process, it can effectively prevent the network from producing local minimum value and local maximum value and other phenomena. According to the experimental trial and error method summary, the momentum factor is generally taken as about 0.85; therefore, this experiment shows that the neural network model with the momentum factor of 0.9 in the trial and error method can achieve the best experimental effect2E.
After the improved BP neural network training is completed, some test data are randomly selected to test the BP neural network model, and the corresponding English language teaching information of the Internet of Things is obtained. The actual value of the overall analysis is basically consistent with the expected value, and there is no large change, which also reflects the reliability of the neural network model. And the data collected is reasonable. The actual value is tested and output by the BP neural network, the actual value is valid, the expected value is reasonable, and the maximum relative error of the two is 1.64%, which basically tends to be ideal. According to Table 5, the actual value of the neural network after training is basically the same as the expected value, but the local fluctuations occur at the 9th, 10th, 15th, 16th, 19th, and 20th sequence numbers, which are also within the acceptable range. Figure 3 shows the comparison between the actual value and the expected value predicted by the neural network.
In the future, education management departments, schools, communities, units, teachers, students, etc. can all be connected through big data. School-school, school-teacher, school-student, teacher-student, school-community, and student-community relationships are like vehicles on the Internet, are these relationships interconnected in the teaching network.
As mentioned earlier, on the basis of the Internet of Things, each industry has new applications; in addition to the Internet of Vehicles, teaching networking can also be expected. “Teaching networking” refers to the use of various sensors to encode and identify all teaching resources and teaching equipment on the teaching site and organize them according to the teaching rules. At the same time, the use of the Internet, wireless communication networks, private networks to publish, according to the user's learning behavior and characteristics of data mining, the use of cloud computing and intelligent information processing to analyze and process data, so as to achieve teaching goals.
4.4. Statistical Analysis of Learning Problems
We screened and dealt with the collected students’ questions and eliminated the students’ problems in course development, production, and equipment, such as course video playing noise, network stutter, and headphone discomfort. After filtering out these questions that have nothing to do with the course Q&A, we will now distribute the number of collected questions and unanswered questions. Based on the statistics of 10 personal questions participating in the experimental study, a total of 56 student questions were collected, of which 52 learning questions were answered and 6 learning questions were not effectively answered. The proportional distribution of the number is shown in Figure 4.
Among the 58 learning questions of 10 learners who participated in the online project course, there were 28 initial input questions, namely, 18 fault operation problems and 10 device principle cognitive problems. After the initial input questions were answered, 9 of the 10 people chose to ask questions again, and a total of 30 questions were reentered: 3 fault operation categories, 16 device principle cognition categories, and 11 project experimental design questions. The statistics of the data distribution of each category are shown in Figure 5.
In the experimental study of this online project course, each student studied two projects: red and green traffic lights and bow alarm, with a total of 20 projects. Among them, 5 projects were completed directly after students took online courses; 13 projects failed to be completed after learning and were completed after troubleshooting with the help of question-and-answer program; and 2 projects were still not completed after students had passed the question-and-answer program. The quantity distribution is shown in Figure 6.
5. Conclusion
On the basis of the Internet of Things, each industry has new applications, such as the Internet of Vehicles. The concept of the Internet of Vehicles is derived from the Internet of Things, which connects cars to vehicle networks through information sensing devices for identification and management. According to the traditional definition, the Internet of Vehicles refers to a system that identifies the electronic tags loaded on the vehicle through sensing technologies such as radio frequency, extracts and effectively utilizes the attribute information and static and dynamic information of all vehicles, effectively supervises the operating status of all vehicles, and effectively provides comprehensive services according to different needs. According to the terminal classification of things, the “things” in the Internet of Things include various forms of “things” such as machines, cars, and elevators. The Internet of Vehicles and the Network of Machines are also included in the application of the Internet of Things in the industry. In the context of big data, objects are constantly connected to each other and to people. To check whether the doors and windows of the home are closed, one just checks the software on the mobile phone and then remotely controls the switch after confirmation. After the Interconnection of All Things, people and people, people and things, and things and things can be connected.
With the development of Internet of things technology, education tends to be more scientific and information-based. The application of network education platform in English teaching has been realized and has broad prospects. The development of intelligent education will undoubtedly cause changes to the traditional English teaching mode. Therefore, when studying the English education system, this paper takes students as the object, deeply understands the relationship between Internet of things technology and intelligent system, and then designs a fully functional English education system. The system not only provides students with personalized teaching services, including online learning, online testing, downloading, and online answering, but also provides a variety of communication channels, for online and offline simultaneous English education, to promote the development of students’ advanced cognitive skills and strengthen mutual learning between teachers and students. The English education system fully guides students to learn and think, increases students’ logical thinking in the process of building a cognitive world, helps them constantly reflect on themselves, and fully stimulates their enthusiasm and initiative. In turn, students’ ability to listen, speak, read, and write English is improved.
Data Availability
The labeled dataset used to support the findings of this study is available from the author upon request.
Conflicts of Interest
The author declares no conflicts of interest.
Acknowledgments
This study was sponsored by Sias University.