社交网络中的抑郁症用户语言和行为特征分析及检测  

Analysis and Detection of Language and BehaviorCharacteristics of Depression in Social Network

作　　者：门秀萍[1] 魏瑞斌[1] 吴小兰[1,2] Men Xiuping;Wei Ruibin;Wu Xiaolan(School of Management Science and Engineering,Anhui University of Finance and Economics,Bengbu 233030,China;School of Economics and Management,Nanjing University of Science and Technology,Nanjing 210094,China)

机构地区：[1]安徽财经大学管理科学与工程学院,安徽蚌埠233030 [2]南京理工大学经济管理学院,江苏南京210094

出　　处：《现代情报》2020年第6期76-87,共12页Journal of Modern Information

基　　金：国家社会科学青年基金项目“社交媒体视域下的跨学科用户发现及其推荐研究”(项目编号:17CTQ047)。

摘　　要：[目的/意义]通过在线工具来分析社交网络用户的语言和行为特征,对抑郁症患者进行预测性研究。[方法/过程]从推特上收集自我披露的抑郁症信息,提取抑郁症用户和正常用户在社交网络中的语言和行为特征信息。在对特征信息进行分析和验证的基础上,利用机器学习算法预测抑郁症用户。[结果/结论]研究发现,用户在社交网络上的语言和行为特征能够反映其心理状态,从中提取的各类特征可用于抑郁用户的检测。在抑郁症用户预测方法中,随机森林分类器的表现最好。基于Empath的词语类别特征在不同类型分类中具有最高的预测得分,而主题特征的得分排名较低。[Purpose/Significance]This study aims to analyze language and behavioral characteristics of social network users by using online tools,and to implement predictive research on depression.[Method/Process]We collected self-disclosed depression information from Twitter,and extracted the characteristics of language and behavior of depression users and normal users in social networks.Based on the analysis of feature information,several machine learning algorithms were used to detect depression users.[Results/Conclusion]The study found that users'language and behavior characteristics in social networks can reflect their psychological state,and the various features extracted from them can be used for the prediction of depressed users.Random forest classifiers performed best in predictors of depression users,the characteristics of empath-based linguistic psychology have the highest prediction scores in different types of classification,while the topic features have lower scores.

关 键 词：抑郁症 社交网络 机器学习 分类 用户 语言 特征分析 检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]