Study On The Identification Of Lonely Students Based On Collaboration Network

Loneliness is a negative emotional experience.Studies have shown that loneliness may cause various problems for college students,such as poor academic performance,mental illness and even suicide.Therefore,it is of great practical significance to identify lonely students in time and accurately.This thesis proposes a method for identifying lonely students based on group collaboration data,focusing on three research work.(1)A method for identifying lonely students based on group collaboration data is proposed,and some novel features are designed.According to the author’s knowledge,no other researchers have proposed a method of identifying lonely people using group collaboration data.First of all,nearly 1,000 course project documents were manually processed,and more than 3,000 group collaboration records were generated.Second,use social network analysis methods to build student collaboration networks,and make visual analysis of the network.On this basis,several novel characteristics considering the characteristics of lonely students are proposed.These features not only help to improve the accuracy of machine learning models,but also help to understand the behavior of lonely students.(2)A oversampling algorithm for graph convolutional neural networks is proposed.The graph convolutional neural network is suitable for the learning of the network structure.There is a serious class imbalance in the data,which leads to the poor recognition effect of the model.This thesis proposes an oversampling algorithm suitable for graph structures.While constructing a few samples,it determines the edges and weights between samples.Experiments show that this method alleviates the problem of class imbalance and improves the model classification performance.(3)An interpretable identification framework and algorithm for lonely students are proposed.Because loneliness may cause various problems and negatively affect students’ lives,loneliness recognition models should not only give classification results,but also give reasons for classification.Due to the low interpretability of graph convolutional networks,this thesis proposes an interpretable recognition framework,including input layer,processing layer and output layer.Experiments show that the newly added features can improve the Recall index by about 16%,and the model can identify more than 83% of lonely students on the real data set.Compared with the scale,the group collaboration data is less subjectively affected by the survey subjects,the data is more objective,and the collection is more convenient.Experiments show that it is feasible to use group collaboration data to identify lonely students.However,this article is only a preliminary study of the method,and its effectiveness needs to be verified in a wider range.