Computational Psychiatric Nursing Research: Scaling up the Prediction of Psychosis by Natural Language Processing
Poster A19, Monday, October 8, 11:30 am - 1:00 pm, Essex Ballroom
Chang Liu1, Quang Hoi1, Ruiwen Xing1, Tran Khanh Trang Kay1, Sunny Chieh Cheng2, Dong Si1; 1Computing and Software Systems, University of Washington Bothell, 2Nursing and Healthcare Leadership Program, University of Washington Tacoma
The prevention of serious mental illness is an urgent public concern due to the burden and cost that mental illness brings to the individuals affected, their families and society at large. Language and speech are the primary data source that mental health professionals use to diagnose and treat mental disorders. This study aims to use machine-learning techniques to distinguish the speech of the patients with schizophrenia from that of healthy individuals to improve the early prediction of severe psychiatric illness. Methods: Forty interviews with patients with first episode psychosis and public messages dataset were included in this study. We applied a well-recognized pre-trained word embedding model that Facebook Research trained on Common Crawl and Wikipedia listings. Word embedding is one of the most popular forms to represent natural language processing because it helps to optimize the accuracy of text data analysis. After the pre-trained word embedding model transfers texts into matrixes, the convolutional neural network is used to develop a classifier, which can distinguish patients from healthy individuals. Results: 80% of each dataset is selected as a training set to train text classifier while the remaining dataset is used as test set. The preliminary results of both test sets achieve a correct prediction rate of 99%. The machine-learning speech classifier achieves to discriminate speech in patients from healthy individuals’ daily conversations. Conclusion: Machine learning technology makes it so the computational models can be trained to learn the features that can predict individuals who will subsequently develop psychosis. This line of inquiry will contribute to improved identification of individuals at risk for psychiatric symptoms as well as lead to development of targeted therapies.
Topic Area: Diagnosis and Phenomenology