A Depression Tendency Detection Model Fusing Weibo Content and User Behavior

Numerous studies on depression have found that tweets from severely depressed users can be used to detect depression. This paper proposes a fusion of text, image, and user behavior model (TPBFM) for depressive tendency detection. To construct user text sequences and user image sequences, TPBFM uses build sequence (BS) and build picture (BP) algorithms to process text and images. TPBFM uses XLNet and ResNet18 networks to extract text and image features. In addition, this paper summarizes and extracts 11 behavioral characteristics from users’ descriptions and posting behaviors, respectively from their Weibo content, social behaviors, and visual characteristics. Finally, the text image and behavior features are aggregated and fed into the recognition layer to achieve the effect of triple model fusion, and then identify the user’s depression tendency. Experimental results show that the TPBFM model can identify depressive tendencies of social network users pretty well. It is superior to single-feature and double-feature detection in terms of precision, recall, and F1, and it has good performance compared with the best methods studied so far under the comparison of equivalent data sets.