基于贝叶斯网络的留守中学生网络成瘾相关因素分析

Analysis of Factors Associated with Internet Addiction among Left-Behind Secondary Students Based on Bayesian Network Model

摘要: 目的：对重庆市渝东南民族地区(石柱县和秀山县)的5290名中学生(留守中学生2553名)进行问卷调查，建立基于bootstrap的贝叶斯网络模型，以期探讨留守中学生网络成瘾行为及其相关因素间的网络关系。方法：单因素χ²、多因素Logistic回归对变量进行初步筛选，使用爬山算法构建拓扑结构并采用极大似然估计法进行参数学习完成贝叶斯网络模型的构建。并通过准确率、敏感度等指标检验模型的精度并将其与Logistic回归模型进行对比。结果：贝叶斯网络模型最终筛选出7个网络成瘾相关的重要变量，与网络成瘾直接相关的因素有3个，分别为上网时长、抑郁和焦虑，与家长沟通情况通过日均上网时长与网络成瘾间接联系。贝叶斯网络模型总体预测准确率为81.60%，灵敏度为93.56%。结论：贝叶斯网络模型能较好地进解释网络成瘾及其相关变量间的依存关系，有助于发现网络成瘾检出的潜在影响因素。

Abstract: Objective: A questionnaire survey was conducted on 5290 secondary school students (2553 left- behind students) in the ethnic areas of southeast Chongqing (Shizhu and Xiushan counties) to build a bootstrap-based Bayesian network model in order to explore the network relationship between left-behind secondary school students’ Internet addiction behaviour and its related factors. Methods: χ² and multivariate Logistic regression were used to initially select the variables, and the construction of the Bayesian network model was established by the hill-climbing algorithm to construct the topology and using the great likelihood estimation method for parameter learning. The precision of the model was tested by accuracy and sensitivity and compared with the Logistic regression model. Results: The Bayesian network model eventually identified seven important variables related to Internet addiction. Three factors were directly related to Internet addiction, respectively, the length of time spent online, depression and anxiety, and communication with parents was indirectly related to Internet addiction through the length of time spent online. The overall prediction accuracy of the Bayesian network model was 81.60% and the sensitivity was 93.56%. Conclusion: The Bayesian network model can provide a reliable explanation of the dependency relationship between Internet addiction and its related variables, and help to identify potential influencing factors for the detection of Internet addiction.