Recognition of Behavioral Disorder From Text Using Natural Language Processing Techniques
No Thumbnail Available
Date
2019-09-09
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Nowadays, a mental disorder is becoming a major challenge to global development. The number of people around the globe who are suffering from one or more mental disorders are growing rapidly. Most of these mental disorders can be successfully treated if they receive the necessary diagnosis and therapy on time.
Psychiatrists use a different mechanism to identify mental illness of their patients such as through providing questionnaire, clinical interview, allowing the patients to write their feeling in the form of narrative and observation. However, providing the proper diagnosis to the patient’s needs sufficient experts. Consequently, developing the system that automatically identifies the disorder from patient narrative text is very crucial to the patients and experts.
Natural language processing incorporation with a machine learning approach is proposed in order to recognize three behavioral disorders from the text: alcohol use disorder, depression, and suicidal thought. The proposed system consists of two phases: training and recognition phase. In the training phase, texts that contain patient text patterns of behavioral disorders are used to create the learned model. These patient texts passed through different operations such as pre-processing, POS tagging and word embedding before fed into a machine learning algorithm. Finally, the multi-channel Convolutional Neural Network (CNN) is used to perform the task of training samples based on identified features of each behavioral disorder with the ultimate goal of creating a learned model. In the recognition phase, the learned model is used to identify the behavioral disorder from the text that is different from the training data.
We have used sentences that indicate each behavioral disorder to create a learned model and test the correctness of the developed system. Each class of behavioral disorder is trained on 85% of the total dataset and tested with 15% of the total dataset assigned for each class. According to the experiment result obtained, implementing the system that considers syntactic and semantic relationship occurred in the patient narrative is an effective method in order to identify behavioral disorder from the patient narrative.
Description
Keywords
Behavioral Disorders, Natural Language Processing, Machine Learning, Convolutional Neural Network, Patient Narrative