Deep Learning-based Pre-diagnosis and Analysis of Psychological Disorders

Abstract

At present, due to the lack of sufficient treatment institutions as well as professional psychotherapists, numerous patients with mental illnesses do not receive help from professional psychiatrists, thus worsening their conditions. In order to determine the condition of mental illness patients at an early stage, this paper applies natural language processing technology to the psychological field and proposes a text-based and deep learning model for mental illness recognition. The BERT (Bidirectional Encoder Representations from Transformers) pre-trained language model is used to complete the sentence-level feature vector representation of mental health text data, and the obtained feature vectors are subsequently targeted and input to a classifier for classification, which can effectively identify depression, anxiety, post-traumatic stress, and unmet mental illness multiple classifications, breaking the previous common depression identification, improving from the original simple depression dichotomous classification to a mental illness multiple classification task, and focusing on a few focal points, which not only saves human and material resources, but also can achieve twice the result with half the effort. Finally, the algorithm is validated using mental health text dataset, and the experimental results show that the lowest F1 value of the trained model on the test set is 0.77, which can achieve fast screening of text content with the tendency of mental illness, reduce the expert labeling workload, improve the labeling efficiency, and provide a new idea for the recognition of mental illness.