Application Of Educational Search Time Series Data Based On ML And CN

With the development of network communication technology,education-related searches have generated massive time-series data.These data reflect the public views and emotional tendencies on education.It is of great significance for Chinese education business to effectively mine the information in the data.Education-related search time series data has the characteristics of large scale,high value,and rapid growth.Different spatial data are mutual independent,heterogeneous and sparse due to the long feedback time.Nowadays,the rapid development of machine learning(ML)and complex network(CN)research has achieved great success in many fields,providing effective solutions for the analysis and prediction of educational time series data.This paper mainly focuses on the trend analysis and prediction of search time series data in specific educational application scenarios based on machine learning and complex network analysis methods.Specific research content can be divided into three aspects:First,a spatio-temporal analysis framework of multi-source data is proposed.It integrates network search data and official government data,and can better discover the feature correlations in time series data.Using the complex function method,entropy weight method and K-means++ algorithm,a solution with sensitive and timely feedback is proposed to quantify the impact of public attention on government education investment in Sichuan Province.It effectively avoids the problem of data sparsity and insularity caused by data barriers,provides a simple and generalizable model for macro education policy effect assessment and prediction,and supports decision-making institutions to utilize network public response more efficiently.Secondly,this paper proposes a time series data analysis framework based on complex theory and Transfer Entropy(TE).It integrates network search data and business data to effectively analysis long-term and dynamic causal relationships and influences within complex networks.By analyzing the information flow of urban-level school district housing in China,the results show that the policy reform has a significant but short-lived impact on the housing information system of Chinses urban school districts.The network search data has a direct causal impact on the school district housing premium.This work provides a useful method to extract public information by utilizing the Internet data for government and business decision-making.Finally,a Chinese parent education anxiety analysis framework and a pilot index based on machine learning are proposed,which integrate network time series data and graph data to obtain large-scale predictive characteristics.By comparing the prediction ability of seven machine learning prediction models on the parent education anxiety index,namely Categorical gradient boosting(Catboost),Adaptive boosting(Adaboost),Extreme gradient boosting(XGBoost),Light gradient boosting(Lightboost),Random forest(RF),Linear regression(LR)and long short-term memory network(LSTM),the PCA-LSTM model with the best prediction effect is obtained,and the model is analyzed and interpreted according to feature importance by XGBoost-SHAP model.Experimental results show that proposed analytical model provides a short,medium,long period trend on the updateable sequence dataset and obtains large-scale public psychology and emotions,which provides support for real-time dynamic analysis of educational equity and early warning information for public opinion risk prevention and control of educational equity...