Behavior of different types of English learners tends to follow different patterns and characteristics, and the analysis of behavioral data is one of the directions for improving English learning and teaching. This study designs a set of behavioral analysis methods based on machine learning for English teaching and learners. The learners’ behaviors are firstly operated with feature extraction and quantification, and the behavioral data are clustered by using the systematic clustering method (HCM) to improve the SOM model. 1DCNN is used to process the learning time-series data and enhance the data mining and performance prediction ability by BiLSTM and attention mechanism, respectively. This paper distinguishes five categories of English learners, such as excellent, diligent, average, procrastinating and negative, and filters out the factors that are highly correlated with English performance, such as the download of learning resources and the number of times of teaching viewing. Comparison experiments show that the ACC of this paper’s achievement prediction model = 0.53, which is better than other comparison methods. Therefore, the idea of this paper based on machine learning methods to analyze the behavior of English teaching and learners has feasibility.

In this paper, some important algorithms in the field of target detection and tracking are optimized. Firstly, Gaussian modeling is performed in the color space for the dynamic background, and the priority is set for ranking. Then introduce adaptive Gaussian component number mixing, adaptively change the weights, and adaptively change the number of mixed Gaussian components according to the pixel color change in the scene to improve the convergence speed of the complex scene. Finally, Kalman filtering and mean drift algorithm are combined to ensure the robustness of target tracking in complex scenes. The single-frame detection time, accuracy, and average tracking error of the algorithm designed in this paper are examined on the dataset, and it is found that the time consumed by the algorithm in this paper in the three scenarios is 242ms, 323ms, and 274ms, respectively, with the highest accuracy of 96.9% and the average tracking error of only 1.5 pixels. The optimization algorithm designed in this paper is able to adapt to the slight disturbance of the background scene and overcome the influence of noise and ambient lighting, which is a target detection and tracking algorithm with good robustness.

The service efficiency of intelligent customer service robots affects the service operation efficiency of enterprises and plays an important role in maintaining customer resources. This paper applies multimodal interaction technology to intelligent customer service system, takes multimodal big language model Qwen-VL as the core, proposes a two-stage relationship multimodal relationship extraction framework based on big language model, realizes multimodal relationship extraction with the help of high-quality auxiliary knowledge, integrates dynamic semantic features and static structural features to complete the multimodal emotion polarity prediction, and constructs multimodal retrieval Q&A system to improve the performance of smart robot performance. Applying the intelligent customer service system in this paper for service practice, the conversation between the intelligent customer service robot and the customer usually ends in about 50 rounds, and the service efficiency is relatively efficient. In the face of customer emotional sentences labeled as happy, complaining and angry, the recognition accuracy under multimodal sentiment analysis is greater than 99%, and the behavior of “notification” and “confirmation” service behavior accounts for the largest proportion of behaviors, and the number of behaviors reaches 560,365 times, 365976 times, which is in line with the expected service behavior of intelligent customer service robots.

This paper carries out a research on the quantitative evaluation of classroom behavior based on the maximum information entropy model, explains the theory related to information entropy and analyzes the concept of information entropy. The improved iFIAS interactive analysis system was used as the main analysis tool, and S-T analysis and time series analysis were used as auxiliary analysis methods to analyze the classroom teaching behaviors. Classroom teaching behaviors were coded and sampled, from which classroom teaching behavior related data were obtained, through which teaching behavior information entropy, redundancy, interaction mode, teaching mode and behavior category frequency were analyzed. After analysis, the overall classroom teaching characteristics of the psychoeducational quality lesson examples are teacher behavior and student behavior as the main, psychoeducational and teaching materials as the secondary, and the teacher’s behavior in the early part of the classroom accounts for a high proportion in order to drive the students into the classroom. The proportion of student behavior rises in the middle and late stages. The proportion of hybrid and dialogic teaching mode is 95.83%, which is dozens of times more than the proportion of lecture. It reveals the teaching mode of quality psychology classroom teachers, i.e., focusing on the interaction with students and replacing the pure teacher lecture student acceptance mode with interactive counseling student active learning. The teaching analysis of quality classroom teaching behaviors with the maximum information entropy model realizes the establishment of an innovative model of psychological teaching and clarifies the direction for the future development of psychoeducational classrooms.

With the rapid expansion of the Internet and e-commerce, and the rapid revolution of the consumption mode, customer reviews have become the most important feedback means for customers’ preference and satisfaction level of products nowadays. In this paper, hotel customer reviews are used as the basis for predicting hotel customer satisfaction, and the TF-IDF feature word extraction method is proposed to extract review feature words. Based on deep neural network, we propose the sentiment analysis technology of hotel customer reviews, use BERT neural network to construct aspect term extraction model, realize the sentiment recognition and quantification of hotel customer reviews, and combine the fuzzy comprehensive evaluation and IPA analysis as the prediction and analysis model of hotel customer satisfaction. Taking 25837 customer reviews of XC Hotel as a research sample, we explore and analyze the satisfaction of XC Hotel customers. The secondary and primary features of the reviews were extracted by review feature words, and the themes were extracted by LDA theme mining model, which concluded that the evaluation items of concern for XC Hotel lie in location, facilities, hygiene, service, price, and food and beverage. The prediction results showed that 69.59% of customers were satisfied, 18.42% felt average and 11.99% were dissatisfied. IPA analysis of satisfaction and importance of XC Hotel and its visualization were conducted, and the intelligent service management model of the hotel was constructed based on the results of IPA analysis, and the optimization strategy of intelligent service of the hotel was proposed.

Variable speed running training method is an efficient training method targeting the improvement of athletes’ agility. In this paper, 30 male students from two badminton special classes of physical education majoring in a college of 2021 were selected as experimental subjects, and hexagonal ball reaction test, hexagonal jump test, repeated straddle test, standing bench press test, closed-eye in-situ step test, and low gravity center of gravity quadrangular running test were chosen as the evaluation indexes of agility quality. A new high-resolution multi-scale feature fusion network was designed for running stance estimation, and the effects of variable speed running training method and conventional agility training on the agility quality of young badminton players were analyzed. The performance curve of the RHPNet designed in this paper has low convergence difficulty and high recognition accuracy, which tends to 0.83, and performs much better than the LSTM network. The intergroup data after the experiments of the experimental group and the control group show that there are significant differences in the performance of the hexagonal jump test, the 20s repeated straddle test, the hexagonal ball reaction test, and the closed-eye in-situ step test. It verifies the effectiveness of the network designed in this paper in the estimation of athletes’ movements during running, and also shows that the training effect of variable speed running training is better than that of conventional agility training.

Disease prevention has always had an important impact on the development of human life and health. The integration of complex network theory and disease has become one of the major trends in epidemiologic research. However, aspects such as individual vaccination behavior and vaccination costs are affected by social capital investment. Based on this, the article investigates a reinforcement learning model of social capital investment on disease prevention. Based on the mechanism of infectious disease dynamics on complex networks, the article investigates the Markov decision process and composition of the reinforcement learning model, and utilizes the theory related to infectious disease dynamics and reinforcement learning to study the mechanism of voluntary vaccination based on epidemic perception. It was found that when the ratio of two kinds of investment (partial investment and full investment) reaches the set maximum value, the full investment policy of targeting selection is more effective in reducing the scale of disease infection in the whole social network and reducing the total social cost, followed by partial investment, and the full investment policy of random selection brings the smallest effect. However, the results may differ for different population investment ratios and partial investment ratios, and both the full investment policy and partial investment policy can effectively control disease prevention, which is conducive to the healthy and prosperous development of the whole society.

Virtual reality technology, as a trend of the development of the new era, has a profound impact and influence on traditional art design teaching. This paper combines virtual reality technology to construct a classroom teaching interactive analysis system to help art design teaching reform. All the objects required for the art design teaching scene are modeled in 3Ds Max, the mapping of each object in the scene is beautified using Photoshop, the FBX format file with animation effect is output, and Unity3D is used for the design and development of the VR part. Subsequently, a quantitative coding form for classroom observation, methods and rules for behavioral data collection, and a classroom migration matrix for analysis were designed to analyze the teaching interaction behaviors in the smart classroom classroom from the micro level. The characteristics of the teaching model in the art and design classroom were captured at the macro level based on the S-T analysis. A teaching experiment was conducted in a school’s art design program after the FIAS analysis. The improved art design classroom interaction increased for the blended teaching mode, and the average of the pre and posttest scores of the experimental class applying this model were 74.18 and 84.37 respectively, and there was a significant difference, which was a significant improvement over the control class. This study provides new ideas and methods for the teaching reform of art design majors in higher education institutions.

In this paper, information theory and information metrics are used to obtain an approximate estimation of linguistic information entropy. After that, the binary model of large-scale corpus and foreign language words is established, N-Gram model is constructed, and the information entropy of modern foreign language speech is estimated. Finally, the N-Gram model was utilized to statistically analyze the results of interpreting information loss, comparing the rate of information transfer in foreign language speeches and the subjects’ interpreting performance. The results showed that the phenomenon of information loss was prominent, with many types of loss, high frequency, and serious loss situations. T assertions had 8.61%-18.95% of propositional information loss, 3.0%-7.6% of constituent information loss, and 49.68% of overall loss. The data on the information loss of each language component showed that TPO and SPE presented the most and the least frequency among the 6 propositional information losses, which were 67 and 1 times, respectively. Among the 13 types of information component loss, TFLS presented the highest frequency and TLE and SFLO presented the lowest, with their losses of 55, 1, and 1 times, respectively. In the interpreted material of English speech, the rate of narration was 2.25 words per second and the average rate was 13.45 bits per second. Among the T assertions, numbers S7, S4, and S9 have the highest propositional untranslated rate (21.8%), propositional mistranslated rate (23.5%), and propositional information loss rate (44.5%), respectively; the corresponding lowest values are at S4 (2.7%), S5 (1.8%), and S4 (2.8%).

Target tracking is a fundamental task in the field of computer vision, which has a wide range of applications in real-life video image signal processing. This paper proposes target tracking optimization technique based on the principle of multi-scale convolutional neural network and multitarget tracking algorithm. The basic structure is designed using VGG16 network, the ROI align method is used to reduce the number of features for feature fusion, and the improved Hungarian algorithm is adopted to associate the fused features and obtain the target tracking results. In the tracking performance experiments, the target tracking optimization technique in this paper is more discriminative in terms of extracted features, and also has higher tracking results under challenging factors such as background clustering (BC), scale variation (SV), and out-of-view (OV). As for the target tracking experiments on mobile network video images, the average tracking accuracy and average tracking success rate of this paper’s method are 97.89% and 96.02%, which are better than DS_v2 and FFT16, and the average error between the target tracking results and the target’s actual motion trajectory is 4.12mm, while possessing the smallest error amplitude.