Research On The Method To Identify Individual College Student With High Psychological Risk Based On Graph Neural Network

Recently,psychological healthy of college students has become an important issue for college and for the whole society.Of students with psychological problems,colleges take a special cares of individual college student with high psychological risk.Those students may suffer from serious psychological disorders,and may do harm to themselves,others,and may even commit suicide.Therefore,they can seriously influence the learning,life and safety of others.Psychological scales are commonly used instruments for measuring psychological health.However,there are many disadvantages to apply these scales for the identification of individual college student with high psychological risk.To solve this problem,a method is proposed in this dissertation which is based on graph neural network.The main works are summarized as follows:(1)We analyze the questionnaires from more than 50,000 students with clustering,regression and visualization techniques.It shows that there is some challenges to identify students with high psychological risk by using data mining techniques,but it is promising to outperform the conventional method used in psychological scales.(2)We propose a process to identify individual college student with high psychological risk based on graph neural network.Conventional machine learning techniques are not suitable for non-Euclidean data set.To make full use of the relationship between students,we designed a process based on the characteristics of graph neural network.Experiments based on GCN(Graph Convolution Network)showed that this process outperformed the general method used by psychological scales.(3)We proposed a bipartite graph convolution network(BGCN).On disadvantage of GCN is that it cannot perform inductive learning.Therefore,we propose a new types of graph neural network based on the particular structure of psychological tests,which is named as BGCN.Experiments showed that the performance of BGCN is better than other machine learning models.Compared with conventional computation method used by psychological scales,data mining techniques have some distinctive advantages.This study illustrates the feasibility of applying BGCN for the identification of individual college student with high psychological risk.It has tested the method on several thousands of students.In the future work,it should be tested in a larger scale.