Advanced machine learning methods in psychiatry: an introduction

Qualitative and Artificial Intelligence-Based Sentiment Analysis of Turkish Twitter Messages Related to Autism Spectrum Disorders

BACKGROUND

The aim of our study was to conduct an emotional analysis of Turkish Twitter messages related to autism spectrum disorders (ASD).

METHODS

An emotion analysis was performed using quantitative and qualitative analysis methods on Turkish Twitter messages shared between November 2021 and January 2022 that contained the words "autism" and "autistic."

RESULTS

It was found that 81.5% of the 13,042 messages that constituted the sample of this study contained neutral emotions. The most frequently used words in Twitter messages were autism, a, universe, strong, patience, warriors, and happy. The qualitative analysis revealed three main themes. These themes were: "experiences," "informing society and awareness," and "humiliation."

CONCLUSION

In this study, it was found that Turkish Twitter messages related to autism, which were analyzed using artificial intelligence-based emotion analysis, often contained neutral emotions. While the content of these messages, often shared by parents, was related to experiences, and the messages shared by pediatric psychiatrists and rehabilitation center employees were informative in nature, it was determined that the word "autism" was used to insult, which is outside of its medical meaning.

Qualitative and Artificial Intelligence-Based Sentiment Analysis of Turkish Tweets Related to Schizophrenia

OBJECTIVE

The aim of this study was to qualitatively examine Turkish tweets about schizophrenia in respect of stigmatization and discrimination within a one-month period and to conduct emotional analysis using artificial intelligence applications.

METHOD

Using the keyword 'schizophrenia,' Turkish tweets were gathered from the Python Tweepy application between December 19, 2020 and January 18, 2021. Features were extracted using the Bidirectional Encoder Representations from Transformers (BERT) method and artificial neural networks and tweets were classified as positive, neutral, or negative. Approximately 5% of the tweets were qualitatively analyzed, constituting those most frequently liked and retweeted.

RESULTS

The study found that, of the total of 3406 schizophreniarelated messages shared in Turkey over a period of one-month, 2996 were original, and were then retweeted a total of 1823 times, and liked by 25,413 people. It was determined that 63.4% of the tweets shared about schizophrenia contained negative emotions, 28.7% were neutral, and 7.71% expressed positive emotions. Within the scope of the qualitative analysis, 145 tweets were examined and classified under four main themes and two sub-themes; namely, news about violent patients, insult (insulting people in interpersonal relationships, insulting people in the news), mockery, and information.

CONCLUSION

The results of this study showed that the Turkish tweets about schizophrenia, which were emotionally analyzed using artificial intelligence were found often to contain negative emotions. It was also seen that Twitter users used the term schizophrenia, not in a medical sense but to insult and make fun of individuals, frequently shared the news that patients were victims or perpetrators of violence, and the messages shared by professional branch organizations or mental health professionals were primarily for conveying information to the public.

Using Minimal-Redundant and Maximal-Relevant Whole-Brain Functional Connectivity to Classify Bipolar Disorder

Background

A number of mental illness is often re-diagnosed to be bipolar disorder (BD). Furthermore, the prefronto-limbic-striatal regions seem to be associated with the main dysconnectivity of BD. Functional connectivity is potentially an appropriate objective neurobiological marker that can assist with BD diagnosis.

Methods

Health controls (HC; n = 173) and patients with BD who had been diagnosed by experienced physicians (n = 192) were separated into 10-folds, namely, a ninefold training set and a onefold testing set. The classification involved feature selection of the training set using minimum redundancy/maximum relevance. Support vector machine was used for training. The classification was repeated 10 times until each fold had been used as the testing set.

Results

The mean accuracy of the 10 testing sets was 76.25%, and the area under the curve was 0.840. The selected functional within-network/between-network connectivity was mainly in the subcortical/cerebellar regions and the frontoparietal network. Furthermore, similarity within the BD patients, calculated by the cosine distance between two functional connectivity matrices, was smaller than between groups before feature selection and greater than between groups after the feature selection.

Limitations

The major limitations were that all the BD patients were receiving medication and that no independent dataset was included.

Conclusion

Our approach effectively separates a relatively large group of BD patients from HCs. This was done by selecting functional connectivity, which was more similar within BD patients, and also seems to be related to the neuropathological factors associated with BD.

Machine Learning Methods to Evaluate the Depression Status of Chinese Recruits: A Diagnostic Study

PURPOSE

Traditional questionnaires assessing the severity of depression are limited and might not be appropriate for military personnel. We intend to explore the diagnostic ability of three machine learning methods for evaluating the depression status of Chinese recruits, using the Chinese version of Beck Depression Inventory-II (BDI-II) as the standard.

PATIENTS AND METHODS

Our diagnostic study was carried out in Luoyang City (Henan Province, China; 10/16/2018-12/10/2018) with a sample of 1000 Chinese male recruits selected using cluster convenient sampling. All participants completed the BDI and 3 questionnaires including the data of demographics, military careers and 18 factors. The participants were randomly selected as the training set and the testing at 2:1. The machine learning methods tested for assessing the presence or absence of depression status were neural network (NN), support vector machine (SVM), and decision tree (DT).

RESULTS

A total of 1000 participants completed the questionnaires, with 223 reporting depression status and 777 not. The highest sensitivity was observed for DT (94.1%), followed by SVM (93.4%) and NN (93.1%). The highest specificity was observed for NN (60.0%), followed by SVM (58.8%) and DT (43.3%). The area under the curve (AUC) of the SVM was the largest (0.862) compared with NN (0.860) and DT (0.734). The regression prediction error and error volatility of the SVM were the smallest.

CONCLUSION

The SVM has the smallest prediction error and error volatility, as well as the largest AUC compared with NN and DT for assessing the presence or absence of depression status in Chinese recruits.