用于社交媒体抑郁评估的对比学习方法研究OA

Social media provides a new avenue for early detection of depression.However,there are still two challenges to overcome,namely,jointly considering textual and visual information and extracting common features related to depression from multimodal data.Therefore,a contrastive learning method is proposed for multi-modal depression assessment.This method encoded textual and visual data,extracted data features of different modalities using feature extraction branches,aligned and fused multimodal features through a multimodal fusion module,and utilized the approach of multi-task learning to assist the model in learning the common features related to depression in both textual and visual information.Experiments show that the proposed method helps improve the accuracy of depression assess-ment.