Depression Detection Based On Text And User Networksa

The extensive utilization of social networks presented an opportunity for detecting depression in a timely and cost-effective manner. In that paper, we proposed TUNDD, a depression detection method based on text and user network, which leveraged social network data. TUNDD comprised three modules: the text embedding module utilized BERT to encode post text, the graph perception module applied GAT to explore potential relationships among users, and the classification module combined post text and user vectors, transforming them into higher-dimensional vectors through BiRNN and generating results through softmax. Experimental results demonstrated that TUNDD outperformed five baseline experiments on two datasets, achieving accuracy rates of 86.3% and 84.7%, respectively. Furthermore, ablation experiments revealed that BERT, GAT, and the number of neural network layers in GAT significantly impacted the effectiveness of depression detection.