Financial risk has a greater impact on the operation and development of enterprises, and accurate prediction of financial risk has become an industry demand, so as to better help enterprises avoid possible financial risk. The article establishes an enterprise financial risk prediction model based on the random forest algorithm, and fills in the oversampling of financial data through the SMOTENC algorithm, and realizes the downsizing of financial data by combining with the KPCA algorithm. Based on the enterprise financial risk characterization index system, the financial data of 358 listed enterprises were selected to carry out model validation and application analysis. The accuracy of corporate financial risk prediction based on Random Forest can reach up to 94.17%, and the average value of the overall time efficiency of the model is 0.68%, which is faster than the comparison algorithm in terms of financial data processing capability. Based on the results of financial risk prediction, the changes in corporate profitability, operating ability, solvency and development ability can be analyzed in depth, providing data support for enterprises to formulate preventive measures for corporate financial risk.

The article uses web crawling to obtain public opinion data after the Sichuan Luding MS6.8 earthquake and preprocesses this data. Aiming at the limitations of the traditional LDA topic model, an improved topic model based on LDA, TT-LDA, is proposed. the BERT model is used to encode the public opinion data, and on the basis of the BERT embedding, the BiLSTM model is used for contextualized word representation for deep feature extraction to complete the modeling of public opinion sentiment evolution. Combining the crawled data and the model, we analyze the public opinion after the Sichuan Luding MS6.8 earthquake. Three days after the earthquake, positive sentiment, neutral sentiment, and negative sentiment increase to 488498, 466832, and 516560, respectively, a total of 1471890 sentiment data, and after time evolution, the sentiment polarity intensity increases from -0.178 to – 0.886, indicating that when the official announcement of the number of casualties of the accident is made, the netizens’ negative sentiment fully erupts to show the post-earthquake public opinion sentiment evolution process.

E-commerce classroom teaching is an important means to improve the quality and teaching effect of e-commerce teaching, and effective interaction in teaching is an important carrier of e-commerce teaching classroom activities. This study combines pan-reinforcement learning and reinforcement Q learning algorithms to recognize and analyze speech data in e-commerce teaching classroom, and uses head posture estimation algorithm to recognize interactive behaviors in e-commerce teaching classroom video, and combines the video and speech interaction data to get the e-commerce teaching interactive behavior recognition model. The model is then equipped with web application technology to design a visual analysis system for e-commerce teaching interaction, and the optimization strategy of e-commerce teaching interaction is realized with the assistance of this system. The results of the study show that the interactive behavior recognition model proposed in this paper can accurately identify the interactive behavior of teachers and students in each course of e-commerce teaching. It is also found that after the implementation of interaction optimization strategy in college e-commerce teaching classroom, the frequency of effective interaction behaviors of teachers and students increases from 351 to 391 times, and the meaningless classroom silence time is reduced. And the learners’ cognition of knowledge is also improved under the influence of the improvement of the effect of interactive behavior. The visual analysis system of teaching interaction proposed in this paper based on reinforcement learning algorithm is of great significance for optimizing the effective interactive behaviors of teachers and students in e-commerce teaching and improving the degree of students’ knowledge cognition.

Green economy is an important factor to measure the quality of economic development. In order to explore the current green economy resource allocation, this paper embeds methods such as DEAMalmquist model and Tobit regression model into the study of green economy resource allocation, explores the green economy resource allocation efficiency of 30 provinces in China by constructing a multilevel model of green economy resource allocation, and analyzes China’s green economy during the period of 2021-2023 through the results of the measurement of the Static, dynamic and level changes of resource allocation efficiency. Tobit regression analysis of the influencing factors of green economy resource allocation efficiency is carried out to optimize the current resource allocation based on the influencing factors. The green economy resource allocation efficiency increases year by year in 2021-2023, and the resource allocation effect improves continuously, with the mean value of the comprehensive efficiency of 0.712, 0.762, and 0.809, respectively. The green economy resource allocation efficiency in Beijing, Shanghai, Jiangsu, and Zhejiang is the highest, and the allocation structure is the most reasonable. Chongqing, Gansu, Qinghai, Ningxia and Xinjiang are less efficient in green economy resource allocation. The per capita GDP and the ratio of education expenditure to GDP have a positive impact on the effect of green economy resource allocation, with an impact of 1.246 and 0.489, respectively.

In recent years, with the rapid development of information technology, the traditional single-threaded processing method can no longer meet the rapid growth of digital media data volume. In this paper, based on the digital media data processing system based on BS structure, the GPGPU parallel processing architecture is used for optimization. The access efficiency of massive parallel multithreading is ensured by executing a multilevel storage architecture composed of behavior decision unit, branch merge unit and branch recovery stack. The study designs the computational resource pool as well as the storage resource pool to form an infrastructure solution to the data processing problem. The query performance of the digital media data processing system using the GPGPU microarchitecture with multithreaded parallel processing is improved by about 81% and 69% or so compared to the Ocelot and prototype systems, respectively. And the average execution time for performing dynamic data allocation is 5.17s less than that of the original system. It shows that the optimized digital media data processing system has better data processing efficiency.

The problems of debt value and optimal capital structure of enterprises are the main issues in corporate finance research. Under the ESG rating mechanism, the article first utilizes the real option theory to study the optimal capital structure and investment and financing decision-making methods of enterprises. Then it puts forward model assumptions and combines the jump diffusion model for the construction of enterprise project investment and financing decision-making model and the dynamic planning adjustment of capital structure. Finally, through specific numerical experiments, the influence process of each variable in the model on the enterprise investment and financing decision is analyzed, and the agency problem is analyzed.Through the experiment, it can be obtained that when the residual value after stopping production, the risk-free interest rate, the variable production cost and the tax rate are set to γ = 1, r = 0.1, ξ = 0.1, θ = 0.2, respectively, with the increase of the frequency of the jump, the investment price of the positive-jump model gradually decreases, and the investment price of the negative-jump model gradually increases, which can be obtained that the reasonable simulation estimation of the relevant parameters has an important impact on the enterprise’s investment strategy, so the enterprise should make a more accurate assessment of the parameters, otherwise they will lose part of the benefits or lose good investment opportunities.

Analyzing the interaction strategies of international relations helps to understand and predict changes in the international landscape, so as to develop and optimize international interaction strategies. Firstly, a single-layer multi-temporal network is modeled for political events, scientific cooperation and international trade in international relations, and added to a multi-layer aggregation network. On this basis, a simulation and analysis method for simulating international relations interaction strategies based on deep learning and multi-intelligence body reinforcement learning methods is proposed. Applying the method of this paper to the arithmetic simulation analysis, it is found that the international relations in the last 10 years have shown the small-world characteristics, and cooperation and conflict coexist. Economic dependence is an influencing factor of conflict between two countries, when the economic and trade links are close, the two countries are less prone to conflict, so the optimization of the international relations interaction strategy should focus on the economic and trade relations.

Objective and comprehensive teaching effectiveness assessment is a strong guarantee for the quality of English teaching in colleges and universities. This paper establishes an index system for evaluating the teaching effectiveness of English in colleges and universities from the levels of students, teachers and classroom teaching. The particle swarm algorithm is used to optimize the convolutional neural network, the SGD formula is used to improve the calculation accuracy, the Adam optimizer is improved to improve the model operation efficiency, and the optimization algorithm of convolutional neural network, PSO-CNN, is proposed. The PSO-CNN algorithm is introduced into the system and the logic design is carried out to realize the evaluation of the teaching effect of English teaching and to build up the English teaching effect evaluation system based on the intelligent algorithm. Evaluation system based on intelligent algorithm. The performance of the system is examined and analyzed with the help of PCA method, which shows that the cumulative contribution rate of the first six indicators, such as learning acquisition and teaching ability, reaches 91.08%. In the mean square error of model training, the PSO-CNN algorithm applied in this paper’s system has a lower mean square error than other algorithms after 35 iterations, and has better evaluation accuracy. In the application practice of English majors in a higher education institution in B city, after applying the system of this paper to evaluate the effectiveness of English teaching, the English test scores of the regular and experimental classes with improved teaching were significantly improved.

Traditional learning path planning methods often fail to meet the individualized needs of learners. In this paper, a dynamic learning path planning method based on neural network is studied by constructing a student model. Firstly, the construction method of the student model is designed, and the Item Response Theory (IRT) is used as a test method for the cognitive level of students, which realizes the dynamic acquisition of student information. A neural network-based cognitive collaborative filtering model was constructed, which models learners’ learning behaviors and interests, and customizes personalized dynamic learning paths for learners after assessing their cognitive levels and learning difficulties. The collaborative filtering algorithm in this paper performs better than the other four algorithms in terms of accuracy and coverage, and the accuracy and coverage rate of the generated knowledge point sequences reach 98.9% and 93.6% respectively, and the performance of the students in the experimental group has been significantly improved under the application of the dynamic learning path generation model of blended teaching in this paper, indicating that the effectiveness and feasibility of the personalized learning path generation model in this paper are excellent and are expected to be further promoted.

The monitoring of training load and recovery cycle of Wushu Sanshou athletes is a long-term and fundamental work for sports teams. The article introduces the parameters of resting heart rate, ventricular muscle contractility, arterial wall and maximal oxygen uptake VO2max as monitoring indexes, and designs a real-time monitoring method of physical training load data based on graph convolution network. Subsequently, through the flow level variables (BFL, TLQ, BRQ), flow rate variables (BFLI, BFLD, TLQI, BRQI), auxiliary variables (TT, TI, RT, RM), exogenous variables (RYN), and the causal relationship between the elements of each variable of the Wushu sparring training function monitoring system, we constructed a nonlinear system of the training load and recovery cycle of the Wushu sparring athlete Dynamics model. Using the real-time monitoring model of this paper to monitor the wushu sparring athletes, in the third minute of the experiment, the real-time monitoring system predicted that the heart rate was 90, and the adjusted heart rate using the model of this paper was 90, which was consistent with the actual monitored heart rate. It can be concluded that the model of this paper can well monitor the training load of martial arts sparring athletes. Through experimental simulation, the article concludes that both the strong physical fitness program and the strong training program can be beneficial to the training of wushu sparring athletes.