Research On Academic Risk Warning And Video Recommendation Methods Based On Learning Behavior Analysis

With the large-scale popularization of "Internet plus education",intelligent education mode supported by information technology such as artificial intelligence has become the trend of educational informatization development.Therefore,personalized learning has attracted extensive attention of researchers.However,online learning lacks effective supervision from teachers and has a large number of resources,making it easy for learners to experience"knowledge loss",making it difficult for learners to continuously follow up on course progress and having a high rate of giving up halfway.In view of this,the thesis takes academic risk prediction as the premise and explores an academic warning and learning video recommendation method based on learning behavior analysis,aiming to track learners’ learning status in real-time and provide corresponding warning and personalized intervention for learners with academic risk.The main work and contributions of the thesis are as follows:Firstly,a learning behavior analysis and convolutional neural network based academic risk warning method was designed to address the issues of single classification results and low feature extraction efficiency in existing warning methods.This method uses Pearson correlation coefficient and Gini coefficient to calculate the correlation between learning behavior and grades,screen out effective learning behavior features,and construct a behavior sequence matrix based on this,which is input into the convolutional neural network to predict learners’grades.Then,four warning levels are divided based on learning grades,providing corresponding warning and personalized suggestions for students in different risk intervals.Secondly,aiming at the problem of insufficient data modeling of learning behavior and sparse scoring data in existing learning video recommendation methods,a learning video recommendation method based on implicit feature migration and multi-step knowledge reasoning is designed.On the one hand,this method constructs a learning interest model and a neural cognitive diagnostic model,quantifying learning interest and cognitive level as learners’implicit ratings of videos and transferring them to the target rating matrix,filling in the sparse video rating matrix,and then using Pearson correlation coefficient to calculate learners’ rating similarity.On the other hand,mining the semantic relationship between video knowledge points,using multi-step knowledge reasoning algorithm to embed all entity relationships into the low dimensional vector space,calculating the semantic similarity between videos,and combining the two similarity degrees of video scoring and video semantics,on the basis of the improvement of collaborative filtering algorithm,generating video recommendation list.The experimental results show that the video recommendation method designed in the thesis has improved accuracy,recall,and comprehensive indicators.Finally,in order to verify the effectiveness and rationality of the aforementioned academic risk warning model and personalized learning video recommendation model,the thesis analyzes the personalized needs of learners from the perspective of practical applications,designs and develops a personalized learning system.This system includes two core modules:academic risk warning and personalized learning video recommendation.It is an application of two models for real scenarios,which can timely identify learners with academic risks and conduct academic warning and personalized video recommendation.In summary,the thesis proposes corresponding academic risk warning models and personalized video recommendation models to address the academic risks and personalized needs in smart education.Through a large number of simulation experiments,it has been proven that the academic risk warning model can accurately identify learners’ academic risks and provide corresponding warning prompts.At the same time,the learning video recommendation model can effectively alleviate the sparsity of scoring data,providing new ideas for the research of academic risk warning and personalized learning video recommendation.