南京大学学报（自然科学版）2025，Vol.61Issue(5)：879-890,12.DOI:10.13232/j.cnki.jnju.2025.05.016
一种基于正无标记学习的大学生心理健康预测方法
A positive unlabelled learning-based mental health prediction method for college students
摘要
Abstract
Traditional manual surveys and interviews have proven effective for mental assessment and screening among college students,yet they require substantial human and material resources.Benefiting from the massive data generated by campus informatization,machine learning techniques have garnered increasing attention from educators and researchers for predicting student mental health.However,students'intrinsic mental states are highly concealed,making sample labeling particularly challenging-especially for negative samples(i.e.,students without mental health issues),which are difficult to define.Previous studies utilized evaluation results from psychometric scales(e.g.,Symptom Checklist-90)as sample labels for modeling,but the highly subjective nature of these scales compromises the authenticity and reliability of conclusions,significantly impairing model performance.To address the critical challenge of defining negative samples caused by the extreme concealment of students'true mental states,we propose PUMPS(Positive Unlabeled Learning for Mental Prediction of Students),a mental health prediction method based on PU learning(Positive-Unlabeled Learning).This approach innovatively introduces the PU learning paradigm and integrates multi-view student-related data to construct a collaborative training framework with multiple ensemble classifiers.By leveraging complementary information across diverse data views,the model's performance is enhanced.Experimental results on real-world datasets demonstrate that the proposed method outperforms traditional approaches across four key metrics:accuracy,precision,recall,and F1-score.Notably,it achieves a 20.73 percentage point improvement in recall over the second-best method.关键词
心理健康/正无标记学习/大学生/半监督学习/多视图学习Key words
mental health/PU learning/college students/semi-supervised learning/multi-view learning分类
信息技术与安全科学引用本文复制引用
盛诗敏,于德湖,马玉玲,周秀艳,聂秀山,李振,张丽梅..一种基于正无标记学习的大学生心理健康预测方法[J].南京大学学报（自然科学版）,2025,61(5):879-890,12.基金项目
国家自然科学基金(62177031),山东省自然科学基金(ZR2021MF044),2023年度教育部人文社会科学研究专项任务(2023JDSZ3174),山东建筑大学国内访问学者经费 （62177031）
