The Study Of Depression Analysis Based On Deep Learning And Social Text

Depression is a common psychological disorder and is the second leading cause of disability worldwide.Traditional depression diagnosis requires communication with the patient,as well as subjective cooperation from the patient,which consumes a lot of human,material,and time costs.With the advent of social texts and the development of natural language processing,computer-aided diagnosis is achieved to provide better and more objective analysis and offer a new way of thinking about depression diagnosis.Therefore,this paper is based on deep learning for deep mining of social text data to establish a model of social text associated with depression and to analyze and study depression.Firstly,the social text is short textual data with unstructured,large volume and limited professional information.It is challenging to construct a research dataset and requires introducing a more significant number of relevant data.As social platforms are aggregated,their reply texts contain relevant information and data augmentation.Secondly,the pre-trained language model BERT encodes social texts,using pretraining as the core idea,combined with feature engineering concepts to optimize the traditional classification task.By introducing a self-attentive mechanism and multiinstance learning,we complete automatic feature selection,enhance feature information,obtain more accurate text vectors,and apply unlabeled data to a supervised task to improve the model’s classification performance.The traditional multi-instance learning algorithm is optimized to be more generalizable by iterative labelling.Finally,the model is applied to a publicly available professional Chinese sentiment rescue dataset for detection,hurtfulness,and attribution classification tasks,using contemporary mainstream pre-trained language models as baseline models,and accuracy,recall,precision,and F1-score as evaluation criteria to verify the classification performance of each component.Each model component is compared and experimented on each task to verify the effectiveness of the model design.This dataset and the publicly available English dataset are analyzed by a word cloud tool to see the similarities and differences between depressed and non-depressed patients from Chinese and foreign cultures,to validate the feasibility of using social media for depression analysis,and to provide a computer-aided diagnosis of depression for psychology and medical professionals.