心理学报2025，Vol.57Issue(11)：2001-2021,中插33-中插34,23.DOI:10.3724/SP.J.1041.2025.2001
数智时代工作紧张人群阈下抑郁的影响因素:基于机器学习的证据
The factors affecting subthreshold depression for people with occupational stress in the era of digital intelligence:Machine learning-based evidence
摘要
Abstract
Depression is one of the most common psychological problems,and subthreshold depression,as a precursor to its occurrence,plays a vital warning role in the prevention and treatment of depression.However,there is currently a lack of in-depth analysis of the representations and influencing factors of subthreshold depression in people with occupational stress in China(SDPOSC).This study integrated grounded-theory research with machine learning methods to explore the manifestations and influencing factors of subthreshold depression among Chinese working population under stress.BERT technology was utilized to construct a discriminant model for identifying the factors influencing subthreshold depression within this population,and the model's effectiveness was subsequently studied and confirmed. This research is composed of two studies.The first study involved the analysis of network texts harvested through web crawling,employing grounded theory for coding to establish a framework of factors influencing subthreshold depression in individuals under stress.The correlation structure between influencing factors and their representations was further explored,along with association rule analysis between influencing factors.Word frequency analysis and occupational difference tests were then conducted to analyze the characteristics of the influencing factors.Mann-Kendall test was subsequently applied to analyze the development trend of influencing factors.Based on the analysis of online text,the second study constructed a machine learning model using BERT technology.the influencing factors of subthreshold depression are distinguished and the effectiveness of the model is subsequently confirmed. Results showed that(1)Manifestations of subthreshold depression in people with occupational stress have five categories,with weakened willpower the highest frequency of expression and daily behavioral changes the lowest frequency.(2)Main influencing factors consist of eight categories,with work factors,evaluation adaptation,and autonomous selection the highest frequency;and stress events the lowest.(3)Eight types of influencing factors were closely related to subthreshold depression symptoms,with stressful event the best single predictor.Network analysis based on association rules revealed that"self-awareness,""behavioral freedom,""environmental adaptation,"and"general social interaction"are the most important subcategories.(4)Healthcare professionals had a significant difference in somatic factors compared to other professions,identified by a difference test of word frequency distribution.(5)Words related to work factors has shown an upward trend from 2011 to 2023,while those related to interpersonal factors have shown a downward trend.(6)A BERT-based machine learning model is obtained and it works in identifying influencing factors of subthreshold depression in populations experiencing work-related stress,in particular,the XGBoost algorithm achieved a prediction accuracy of 81.58％,with particularly strong performance in subthreshold depression detection(F1-score=0.90,AUC=0.93). This study provided an in-depth analysis of the representations and influencing factors of SDPOSC,enriching the localization research of subthreshold depression from an empirical perspective.Furthermore,a machine learning model by BERT can be utilized in subsequent research.The study of SDPOSC can help identify their depression risk and has important theoretical and practical significance for the prevention and treatment of SDPOSC.关键词
阈下抑郁/工作紧张/机器学习Key words
subthreshold depression/occupational stress/machine learning分类
基础医学引用本文复制引用
邓丽芳,裴蓓,高天艾..数智时代工作紧张人群阈下抑郁的影响因素:基于机器学习的证据[J].心理学报,2025,57(11):2001-2021,中插33-中插34,23.基金项目
北京自然科学基金面上项目(7202101). （7202101）
