Cross-language text categorization techniques can achieve more efficient localization and use of text data in multilingual languages by overcoming the differences between different languages. In this paper, firstly, by combining cross-language word vectors and adversarial training, support vector machines are utilized to improve the alignment effect of English-Chinese cross-language words and sentences in the feature space, and to achieve higher quality English-Chinese cross-language text classification. Then the variational mechanism is combined with multi-task learning to align the potential semantic space of multimodal data, maintain the domain invariance of different modal data representations, improve the generalization ability of the model, and ensure the consistency of the variational machine translation training process and the prediction process. The two are combined to construct a hybrid variational multimodal machine translation model based on semantic alignment, experimentally validate the effect of the text categorization algorithm on datasets such as Multi30k, and examine the quality of English-Chinese and Chinese-English translations. In the experiments, it is found that on the MSCOCO dataset, the BLEU of English to Chinese and Chinese to English of this paper’s model is 61.26 and 60.15 respectively, and the translation quality is significantly better than the baseline model. The model achieved the best results in all 3 actual translation tasks. And compared with the complete model, the translation performance of different removal cases in the ablation experiments are decreased, which verifies the effectiveness of the model of this paper as a whole and different components. The method in this paper can effectively reduce the feature differences between different languages, and has important practical application value for solving cross-language text categorization and machine translation problems.

As a small-scale power generation and distribution system, microgrid, by virtue of its high efficiency and clean power generation, has been taken by scholars around the world as a key research object for the sustainable development of national energy. Taking microgrid as the main research object, this paper explores the construction of power load identification model and optimization of scheduling capacity of microgrid. The improved Least Squares Support Vector Machine (LS-SVM) algorithm is used to construct the power load identification model, which realizes the accurate prediction of power load data. The optimal scheduling model of the microgrid is constructed based on the nonlinear planning method, and the co-evolutionary genetic algorithm (DCGA) with the improved difference strategy is used to solve and find the optimal model.The curve of the predicted value of the power load of the LS-SVM is basically fitted to the curve of the real value, and its prediction of the power load is more accurate than that of the BP neural network model. The daily running costs of the genetic algorithm, CCGA algorithm and DCGA algorithm are 1750.34 yuan, 1730.59 yuan and 1709.83 yuan, respectively. The daily running cost of the improved DCGA algorithm in this paper is 1763.59 yuan, which is reduced by 2.31% and 1.20% compared with the genetic algorithm and co-evolutionary genetic algorithm, respectively, and the DCGA algorithm has the fastest convergence speed, which indicates that it has the strongest ability to search for optimization, and it can effectively reduce the operating cost of microgrids, and it has a high practical value.

In this paper, the embedding vectors are obtained by Bert coding, and then the obtained embedding vectors are adaptively fused with features to realize legal text classification by a classifier, on the basis of which a multi-label text classification model (AFDAM) is proposed to capture the target words in a sentence. At the same time, the pre-trained continuous bag-of-words representation (CBOW) is used to initialize the vector representation of the label information, and then these label information is adaptively fused with the feature information of the text, which effectively promotes the multi-label legal text classification, and accelerates the development of informationization and intelligence in the legal field. The results show that the text feature enhancement module has the most prominent impact on the text classification effect, and its accuracy on the three datasets is improved by 0.46%-1.19%. In addition, the introduction of target vectors and text expansion also gained 0.54%-1.7% and 0.59%-1.53% and 1.08% increases in model accuracy, respectively. In addition, the addition of offense and statute information can significantly improve the prediction of sentence length, and the statute information improves the results more significantly than the offense information. And the classification effect of the AFDAM model proposed in this paper increased by 0.1453-0.257 than the other five models.

Quality management is one of the factors determining the running level of a university, so it is necessary to evaluate university management scientifically and comprehensively. In this paper, a university management evaluation index system is constructed and optimized with reference to the multi-factor decision tree technology, and the cause degree and centrality degree are calculated through the DEMATEL method model, the causality diagram is established, and the ISM hierarchical structure analysis is subsequently carried out. The management influences with high centrality degree are service ability, student quality, employment, school size and presidential leadership. Through the results of the reachable matrix, four levels of college management influencing factors are divided, and it is found that the fundamental factors affecting the management level of colleges and universities are concentrated in the socio-economic and cultural level, the deeper factors are mainly the school’s own scale, funds and teachers and students, and the leadership quality and service ability are the superficial factors. Therefore, the improvement of university management evaluation system can be carried out with reference to the above levels and indicators.

With the rapid development of social networks, the powerful interactive function of social networks and the high degree of user participation make the information generated in large quantities and spread rapidly, which also provides a dissemination path for the mass dissemination of Marxism. The article analyzes the overall structure of complex social networks and node centrality indexes for the overall topological characteristics of the networks, in order to further analyze the information dissemination characteristics of social networks for the mass dissemination of Marxism. Subsequently, a social network information dissemination model based on SIR is established according to the specific structural characteristics and dissemination modes of the social network, and the experimental test of the information dissemination effect is carried out. Finally, a Marxist mass communication strategy is proposed based on the experimental results. In the experiments on the effect spreading of node information, the nodes numbered 4, 12, and 20 have the strongest information spreading effect, with the number of nodes infected exceeding 30, and the corresponding number of interaction activities are 575, 511, and 663.This suggests that individuals within the aggregated group can build trust with the group members by participating in frequent interactions to improve the effect of information spreading. The development and dissemination of Marxist mass social networks cannot be separated from a series of measures such as a sound social network regulatory system.

This paper introduces a multilayer Bayesian model based on probabilistic and Bayesian inference models to infer discourse hierarchical features in the English corpus at both intra-individual and inter-individual behavior levels. Based on the existing English corpus observation data, the Bayesian method is used to organically combine the prior knowledge and the observed English corpus discourse hierarchical features data, derive and incorporate the posterior probability distribution of many uncertain information target variables, and obtain the discourse hierarchical features of excellent English teachers in the English corpus, which provides more valuable information for educational management. For the emotional features of different teachers’ classroom discourse, the level of “emotionally full” is higher than that of “emotionally depressed”, which indicates that the classroom emotions of excellent teachers are more full, demonstrating the discourse hierarchy features of excellent English teachers. The Bayesian estimation has excellent estimation accuracy and explains the discourse hierarchy of teachers in the English corpus well.

Apriori algorithm is a classic frequent itemset mining algorithm, but it has the problems of more time consumed by the self-connection process and high overhead of conversion between memories. In order to improve the frequent itemset mining effect of Apriori algorithm, this paper improves the existing adaptive genetic algorithm by using the average population fitness and fitness value discretization, and improves the Apriori algorithm by using the optimized genetic algorithm, so as to solve the strong association rules. Compared with the traditional Apriori algorithm, the algorithm in this paper has less time overhead and improves 2.4%, 2.4%, and 2.7% on average in recall, accuracy, and F1 value. On the Accidents and Retail datasets, the improved Apriori algorithm is faster than the NSFI algorithm by 6.12% and 13.52% on average, reducing the computational complexity. Using the improved algorithm to analyze the characteristics of cross-provincial migrants, it is found that the migrant population is younger in age, with lower education level, mostly of agricultural household registration, and mostly of Han nationality, which verifies the practical application value of the improved algorithm.

Teachers’ educational decision-making behavior is a deep factor affecting the quality of teaching and has a guiding role in the whole process of teaching activities. In this paper, lagged sequence analysis is used to focus on comparing the differences in multi-objective educational decision-making behaviors between backbone teachers and novice teachers. At the same time, a collaborative filtering recommendation algorithm based on improved cosine similarity combining teacher users and teaching resources is designed to achieve personalized teaching resources recommendation for teachers. And the personalized teaching path for teachers was designed by combining the characteristics of teachers’ educational decision-making behaviors. In terms of static decision-making behavior, backbone teachers pay more attention to cognitive decision-making, while novice teachers pay more attention to procedural decision-making. In terms of dynamic decision-making behavior, backbone teachers’ decision-making strategies are more balanced and diverse and goal-focused than novice teachers. The personalized teaching path of this paper is much better than traditional teaching methods in actual teaching experiments, and there is a highly significant difference between the pre and post-test scores of students in the experimental group using the path (p=0.000<0.01), and teachers are more satisfied with the accuracy of the resource recommendation and the teaching effect of the path. The personalized teaching path designed in this paper helps teachers' educational decision-making in teaching and provides a feasible implementation path for personalized teaching.

The physical and mechanical properties of the rock body at the foot of the slope are prone to deterioration under water-rock action, which affects the stability of the slope body. Accurate understanding of the damage mechanism of anticlinal rocky slopes in reservoir area under the condition of deterioration of the rock body at the foot of the slope is the key to the reasonable evaluation of stability. In this paper, the main lithological characteristics of the anti-dipping rocky slopes in the reservoir area and the distribution characteristics of slope height, slope angle and inclination angle of the rock layer are investigated as the research object, and the deformation and damage characteristics and laws of the rock body are obtained. Numerical simulation of anticline slopes was carried out using GDEM mechanical analysis software based on the discrete element method of continuous medium mechanics. It is found that the upper and middle parts of the slope where the invert body is located in the studied engineering example have deep tensile cracks and shallow surface block tipping damage, while the middle and lower parts show deep bending deformation, and there is a gradual transition zone in the contact between the deformed rock layer and the bedrock. As the distance between the cave and the basement increases, the rock layers gradually tilt towards the Yellow River from the north-east to the north-west, with the rock layers at the base of the cave tilting between 340° and 350°. The inclination of the rock layer in the example slope is 76°, and the main rupture surface of the slope, i.e. the location of the largest bending moment of the rock layer, has a small inclination angle with the horizontal plane. The slope angle is 48°, and the sum of the angles of the slope angle and the inclination angle of the rock layer is obviously larger than 117°, and the slope will be deformed and damaged, which is consistent with the value of the conditions for the slope to be deformed and damaged.

In recent years, with the increasing psychological pressure on students, psycho-pedagogical methods have been highly emphasized. This article takes students’ multimodal emotion recognition as a research perspective. The article firstly studies the unimodal emotion recognition methods of expression, text and speech respectively. Then it proposes a multimodal emotion recognition algorithm based on dual-attention mechanism and gated memory network, and then conducts emotion recognition experiments to validate this paper’s method. The article further proposes an intervention pathway to further assist in solving students’ mental health problems by designing a virtual reality mental health intervention system. Using the method of this paper in Multimodal database unimodal emotion recognition experiments, found that the network of the model used in this paper has better experimental results, which verifies the effectiveness of the method of this paper, and the accuracy rate of emotion recognition is 60.65%. After testing the mental health level of 8000 students in a school, it was found that the number of hypermodality and the screening rate were low except for the high score of compulsion, from which it can be concluded that the students in our school are in good mental health as a whole after applying the method of this paper.