A Detection Method for Depression Based on Imbalanced Social Media Text
To address the challenges faced by the current depression detection model based on social media data,such as difficulties in handling imbalanced data and incomplete evaluation indicators,we propose a new approach called Document Adaptive Enhanced Bagging-τSS3(DAEB-τSS3).This method utilizes social media text data for depression detection and introduces a novel machine learning evaluation metric called GF(α,β)-Score.Building upon theτ-SS3 model,we incorporate confidence weighting to amplify the influence of certain data types.Additionally,we employ the Bagging method to enhance integrated learning,improving the sampling process from random sampling to layered sampling.This adaptive enhancement focuses on a select number of data documents,thereby improving the model's ability to handle imbalanced data.In the model evaluation stage,-we utilize GF-Score for automatic parameter selection and discard underperforming base learners,thereby enhancing the model's reliability and stability.Experimental results on the E-Risk2017 depression detection dataset demonstrate that DAEB-τSS3 exhibits superior adaptability to imbalanced datasets and outperforms τSS3,bi-directional long-term memory networks,and ERNIE 3.0 models.The average improvements in GF-Score,Fl-Score,and G-Mean Score are 13％,0.7％,and 26.9％,respectively,enabling more effective depression detection based on imbalanced social media texts.
imbalanced datasetdepression detectionensemble learningtext classificationsocial media text data
万方数据
维普
