The application of modern information technology in track and field training has become an important means to improve the training effect. The study analyses the application of smart wearable devices in track and field training, takes the real-time feedback data of smart wearable devices as the index observation point, constructs the evaluation index system of track and field training based on smart wearable devices, and explores the application of factor analysis and fuzzy comprehensive evaluation method. On this basis, teaching experiments are carried out using smart wearable devices and the evaluation system to explore the effect of smart wearable devices on the enhancement of track and field training in athletic performance. The track and field training of the students in the sample colleges and universities was of medium level, with a total score of 73.71, in which the development of students’ will quality and teachers’ grasp of the training situation still need to be improved. After training with smart wearable devices and assessment system, the practicing students got 4.09%~5.01% improvement in standing long jump, 50m run and 800m run, and there was also a significant difference in training interest with the control students (P<0.05). The smart wearable device and evaluation system can achieve real-time data monitoring and training feedback, which can help coaches and students adjust training in time and improve the effect of track and field training.

Writing skills not only promote the learning of other English skills such as listening, speaking and reading, but also effectively promote the internalization of language knowledge, laying the foundation for further improving the development of students’ comprehensive language skills. In this paper, with reference to the application path of information technology in English literacy teaching, we design a SCN-LSTM-based language model, and on this basis, we adopt a bidirectional recurrent network as the language model, and propose an improved SCN-BiLSTM network, which can effectively obtain the contextual relationship of the input sequence. Through the linear interpolation of the language model, the cached language model adaptation is obtained, and the teaching scene corpus is utilized to train the model, and the teaching context-oriented language model adaptation is obtained. Construct ANFIS model to improve the evaluation of English literacy teaching. After the empirical research experiment, the average English reading score of the students in the experimental class after the experiment is 53.631, which is 11.942 points higher than that before the experiment. The writing score is 8.45, which is 0.97 points higher than before the experiment. The application of the adaptive model of English reading and writing based on SCN-LSTM network is very effective.

The era of big data in education has come, data-driven intelligent decision-making has become the development trend in the era of big data, and precise teaching has become the keyword in the era of big data. This paper establishes a real-time dynamic teaching strategy adjustment decision-making model based on the learning characteristics in the process of industry-teaching integration practical training in higher vocational education, and uses Markov decision-making and Q-learning algorithms to solve the optimal teaching strategy in each stage of practical training and learning, which assists the teachers in decision-making and precise intervention. The results of the practical training teaching experiment found that the students in the experimental group, after the dynamic adjustment and intervention strategy implementation of the industry-teaching integration practical teaching, the scores of the practical training theory and application knowledge test were significantly improved (P<0.05), and the students' self-efficacy control sense, sense of effort, and sense of competence were all improved to different degrees. In addition, the scores of depth of understanding (P=0.000) and strategic approach (P=0.000) in practical training learning competencies also increased significantly. The strategy proposed in this study is able to capture the dynamic characteristics of educational data and use the multi-stage dynamic decision-making method to study the development of teaching strategies, which can provide stronger support for accurate teaching decisions and industry-teaching integration of practical training learning.

Prediction of legal decisions using machine learning and artificial intelligence techniques has gradually become an important part of smart court technology. In addition the crime prediction and law recommendation also face the problem of easily confusing crimes. In order to solve these problems, this paper unites multi-task learning models and proposes a model fusion legal verdict prediction model. An attention neural network fusing Transformer Encoder and DPCNN encodes the key semantic information in the case description. The TF-IDF algorithm and TextRank algorithm are applied to extract the keywords of the charge, and the forward propagation network is used as a classifier to constitute a multi-task learning legal verdict prediction model. Using 9 CAIL2018 legal datasets as experimental data, the metrics performance of the multi-task learning legal judgment prediction model proposed in this paper is measured on three subtasks (offense prediction, legal provision prediction, and punishment duration prediction) in LJP. Combining real case information for legal verdict prediction as well as charge differentiation. The verdict prediction results on the CAILBig-Multi dataset show that the mean MP value of the comparison algorithms is 82.925% in the charge prediction. And the MP index of the charge prediction of the multitask learning legal verdict prediction model proposed in this paper is 89.13%, which is significantly higher than the mean value of the comparison algorithms. And the multitask learning model incorporating the keyword information of charges in case analysis can effectively solve the problem of confusing charges.

As the main link of international trade, logistics plays a pivotal role in the entire international trade transactions, and choosing the appropriate logistics path is conducive to cost savings for enterprises. This study combines the traditional logistics model with the actual situation of international trade to select the headway transportation, overseas warehouses and tail distribution as the main elements of enterprise logistics cost optimization in international trade. Based on the cost calculation of the main elements, we design the objective function and constraints of enterprise logistics cost optimization, build the optimization model, and obtain the optimal solution by iterative analysis using the fitness function and genetic operator in genetic algorithm. The empirical analysis shows that after applying the optimization model, the total logistics cost of enterprise D is reduced from US$99,373,500 to US$72,653,400, indicating that the model is effective in optimizing the logistics cost of enterprise D in international trade. This study provides an effective method for the optimization of cross-border enterprise logistics costs, which has a positive role in promoting the development of international trade.

Research on event extraction and constraint encoding of legal cases, using Lawformer as a pre-trained language model for legal sentence prediction model, constructing MJP-Law model to predict the sentence of legal cases. The HAN encoder in the model is utilized to extract the inter-sentence relations in the legal case and construct the relations among the law, the charge, and the sentence period. Compare the performance of this paper’s MJP-Law model with other prediction models on law, charge, and sentence period, and explore the effects of the three subtasks of law, charge, and sentence period on the model through ablation experiments, and compare the prediction effects of a single MJP model and the MJP-Law model on low-frequency charges. In this paper, the MJP-Law model outperforms other prediction models in terms of prediction performance on statute, offense, and sentence. The four models of “MJP-Law”, “MJP-Law_law”, “MJP-Law_SG” and “MJP” had the same prediction performance, which were 95.54%, 89.86%, 89.73% and 89.81%, respectively. “MJP-Law” and “MJP-Law_law”, “MJPLaw_SG” and “MJP” have the same performance in law prediction. After removing the sentencing guidelines and legal sentences, the macro F1 values of the MJP-Law model all showed a decrease.The predictive performance of the MJP-Law model on low-frequency offenses was better than that of the single MJP model.

This paper defines doctor-patient interaction from the perspectives of interaction form and maintenance of patients’ health respectively, and also constructs a doctor-patient interaction discourse model. Based on the data mining technology to obtain the research data, the acquired data are preprocessed and stored in the form of dataset. Bi-LSTM is used to extract topic sentence features from the dataset, and the unsupervised pattern is transformed into a self-supervised pattern through the training and learning of auxiliary tasks to complete the construction of the discourse model of doctor-patient interaction based on topic structure. Combined with the processing flow of natural language processing and semantic technology, the communication strategy generation system for doctor-patient interaction discourse is designed, and finally the communication strategy based on natural language technology is researched and analyzed. There are significant differences between the experimental group and the control group in terms of expression ability and cognitive level (P<0.05), which concludes that compared with the traditional discourse model, the doctor-patient interactive discourse model has a higher priority, and it can effectively improve the expression ability and cognitive level of the patients' medical terminology. On the CMedQA2.0 dataset, the average performance of this paper's model is improved by 46.34% compared with the baseline model GPT-2, indicating that this paper's model has excellent performance. Under the condition of Chinese participle and topic extraction fusion, the average accuracy of this paper's system is as high as 85.02%, which indicates that the system can provide doctors with precise communication strategies based on patients' medical-related information, thereby effectively enhancing the discourse communication skills in doctor-patient interactions.

In response to cybersecurity threats such as security breaches, data leakage, supply chain attacks, and ransomware viruses in digital network environments, more reliable cybersecurity architectures are needed to address these challenges. The article builds a zero-trust firewall applied to network security protection based on zero-trust architecture by integrating SPA single-packet authorisation technology and authentication scheme. Then SPA single packet authorisation technology with SM3 hash algorithm and SM4 algorithm for fully nominal encryption processing is constructed as a network security protection scheme, and the authentication protocol and trust evaluation algorithm are established by using hash and different-or function. In the simulation verification results, the communication volume of SDP client to complete one authentication is 981B, which reduces 27.17% compared to WaverleySDP overhead. The server in the SDP+SPA scenario still retains a certain amount of legitimate data after DDOS attacks and Web attacks, and receives only 53.47% of the traffic of the SDP scenario. The CPU usage of the client deployed with SPA is only 11.47 percentage points higher than that without SPA mechanism. The combination of SPA single-packet knocking technology and zero-trust architecture can achieve network security protection, and can also effectively deal with DDoS and Web attacks, and improve the performance of network security protection.

Consumer data is an important support for analysing and observing consumer behaviours in the era of digital marketing, and constructing models to predict consumer purchasing behaviours. In this paper, we select the Retailrocket consumer behaviour dataset based on real shopping websites, analyse the distribution of various types of consumer behaviour over time and other data characteristics, and gain insights into the behavioural habits of consumers when shopping. Based on the XGBoost algorithm in machine learning, a prediction model of consumer behaviour is constructed, and the genetic algorithm is used to optimize and improve the XGBoost algorithm.The XGBoost prediction model has a significantly better prediction performance than the LSTM prediction model and the LR prediction model when facing the data under the under-sampling data balancing method and the improved random under-sampling method based on the K-means algorithm. . The performance of the GA-XGBoost prediction model optimised by the genetic algorithm is significantly improved compared to the XGBoost prediction model, and substantially better than the LSTM prediction model and the LR prediction model. The accuracy and F1 value of the GA-XGBoost prediction model in the data under the improved stochastic undersampling method are 0.90865 and 0.92435, respectively, which are improved by 14.69% and 17.26% relative to the XGBoost prediction model. Meanwhile, the stability of GA-XGBoost prediction model is also significantly improved compared to XGBoost prediction model.

The digital era requires enterprises to pay attention to technological innovation and optimise ESG performance in the development process, so as to achieve high-quality development. Based on this, this paper proposes the hypotheses related to enterprise ESG, technological innovation and enterprise high-quality development. And construct the regression model of enterprise ESG performance and high-quality development. Basic statistics and correlation analysis are used to provide a preliminary description of enterprise ESG performance and high-quality development. Through the total effect test, the role of enterprise ESG performance on high-quality development is clarified. Through the mediation effect test, the role played by technological innovation between corporate ESG and highquality development is clarified, and the proposed hypotheses are verified, and the property rights, geographic and industry differences in the impact of corporate ESG performance on high-quality development are further explored by using robustness test and heterogeneity analysis. Finally, corresponding recommendations are made. Most of the enterprises selected in this paper have low levels of high-quality development, unsatisfactory ESG performance, and large overall gaps in technological innovation.The correlation coefficients of ESG performance (ESG) with corporate highquality development (LnTFP) and technological innovation are 0.402 and 0.335, respectively, and all of them are significantly and positively correlated at the 1 per cent level. Hypotheses H1, H2, and H3 are all valid.ESG performance and technological innovation have more significant effects on the highquality development of state-owned enterprises, eastern regions, and high-pollution enterprises.