Prediction of Psychological Distress in Persons Through Narrative Writing by Using Natural Language Processing

1. Uluslararası Veri Analitiği Kongresi , İzmir, Türkiye, 29 - 30 Temmuz 2024

Abstract

Introduction and Aim: Today, natural language processing (NLP) methods find applications in

various fields, ranging from language translation to chatbots and document analysis. NLP

techniques are frequently used in fields such as text analysis and classification. The primary aim

of this research is to develop NLP-based algorithms that accurately classify the mental health

conditions of individuals, particularly focusing on mental disorders such as bipolar and

depression, and present the outcomes of these algorithms to enhance the precision of future data

classification. The study seeks to understand psychological states through social media text and

create decision support models for psychologists and psychiatrists.

Method:, This study utilized the data based on the comments of 3368 users received from the

reddit social media platform between 01.01.2020 and 01.04.2020. In this study, classification

models were developed using three machine learning algorithms—naïve bayes, support vector

machines, and decision trees—to distinguish between two labeled classes: bipolar disorder and

depression. To enhance the reliability of all models, 10-fold cross-validation was applied. This

method was used to evaluate the performance of the models and to prevent overfitting. All

analyzes were performed with the R programming language.

Results: Upon analyzing the results, it was found that the decision tree classifier produced the

best performance in disease classification when examining evulation metrcis. The accuracy,

precision, recall and F-1 score of the decision tree model was 0.87, 0.97, 0.72 and 0.82,

respectively. Meanwhile, the support vector machines model had an accuracy of 0.79 and the

naïve bayes model had an accuracy of 0.74.These findings highlight the potential of decision tree

models in accurately classifying mental health conditions.

Conclusion: It is expected that the outcomes of text analyses and models were used in the study

will contribute significantly to future studies in the field of psychology and text classification.The

successful application of NLP methods in this study demonstrates the potential of these methods

to assist mental health practitioners in diagnosing and monitoring psychological disorders.

Keywords: Classificatio, mental diseases, natural language processing, social media