Mao Huimin, Han Qing
School of Medical Technology and Information Engineering, Zhejiang Chinese Medical University, Hangzhou, China.
Front Psychol. 2025 Aug 26;16:1612769. doi: 10.3389/fpsyg.2025.1612769. eCollection 2025.
Depression, also known as depressive disorder, is a pervasive mental health condition that affects individuals across diverse backgrounds and demographics. The detection of depression has emerged as a critical area of research in response to the growing global burden of mental health disorders.
This study aims to augment the performance of TextGCN for depression detection by leveraging social media posts that have been enriched with emotional representation.
We propose an enhanced TextGCN model that incorporate emotion representation learned from fine-tuned pre-trained language models, including MentalBERT, MentalRoBERTa, and RoBERTaDepressionDetection. Our approach involves integrating these models into TextGCN to capitalize on their emotional representation capabilities. Furthermore, unlike previous studies that discard emoticons and emojis as noise, we retain them as individual tokens during preprocessing to preserve potential affective cues.
The results demonstrate a significant improvement in performance achieved by the enhanced TextGCN models, when integrated with embeddings learned from MentalBERT, MentalRoBERTa, and RoBERTaDepressionDetection, compared to baseline models on five benchmark datasets.
Our research highlights the potential of pre-trained models to enhance emotional representation in TextGCN, leading to improved detection accuracy, and can serve as a foundation for future research and applications in the mental health domain. In the forthcoming stages, we intend to refine our model by incorporating more balanced and targeted data sets, with the goal of exploring its potential applications in mental health.
抑郁症，也称为抑郁障碍，是一种普遍存在的心理健康状况，影响着不同背景和人口统计学特征的个体。随着全球精神健康障碍负担的不断增加，抑郁症的检测已成为一个关键的研究领域。
本研究旨在通过利用富含情感表征的社交媒体帖子来提高TextGCN在抑郁症检测方面的性能。
我们提出了一种增强的TextGCN模型，该模型整合了从微调后的预训练语言模型（包括MentalBERT、MentalRoBERTa和RoBERTaDepressionDetection）中学到的情感表征。我们的方法包括将这些模型集成到TextGCN中，以利用它们的情感表征能力。此外，与之前将表情符号和emoji作为噪声丢弃的研究不同，我们在预处理过程中将它们保留为单独的令牌，以保留潜在的情感线索。
结果表明，与五个基准数据集上的基线模型相比，增强后的TextGCN模型在与从MentalBERT、MentalRoBERTa和RoBERTaDepressionDetection中学到的嵌入集成时，性能有了显著提高。
我们的研究突出了预训练模型在增强TextGCN中情感表征方面的潜力，从而提高了检测准确性，并可为心理健康领域的未来研究和应用奠定基础。在接下来的阶段，我们打算通过纳入更平衡和有针对性的数据集来优化我们的模型，目标是探索其在心理健康方面的潜在应用。
