TY  - BOOK
AU  - Ahadi,Seyedeh Aridis
AU  - Jazayeri,Kian
TI  - DETECTING INDICATORS OF MENTAL DISORDER IN SOCIAL MEDIA POSTS VIA HYBRID DEEP LEARNING AND NATURAL LANGUAGE PROCESSING
PY  - 2024///
KW  - Information Technologies
KW  - Dissertations, Academic
N1  - Thesis (MSc) - Cyprus International University. Institute of Graduate Studies and Research Information Technologies
N2  - Identifying people who are suicidal on social media has become more crucial recently. Systems that predict people's mental health can be developed using the information offered by the textual data. In recent times, individuals have turned to social media platforms to share their experiences and seek mental health support. This trend has spurred researchers to utilize the data, applying various Natural Language Processing (NLP) and Machine Learning (ML) techniques to assist those in need. In this study, a framework to identify suicidal thoughts in social media using a combined Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) classifier model is proposed. Different combinations of embedding, activation functions, and solver algorithms are implemented on this network, and then the results are compared. The used methods contain Count and TF-IDF as word embedding, RReLU, Tanh, Mish, and ELU as activation functions, and Adam, Adamax, Adadelta, Adagrad, and RMSprop as solver algorithms. Bagging and stacking classifiers have been utilized on the network to create an ensemble of the basic models. In total, 82 different methods have been applied. The main dataset is gathered from four Reddit mental health datasets on suicidality. Overall, 59996 user texts from 2018 to 2020 were retrieved to be analyzed. Results are compared based on 5 performance metrics (accuracy, F1 score, precision, recall, and AUC) and elapsed times. The output accuracy scores are in the range of 74% - 86%. The combination of TF_IDF, RReLU, and Adam achieved the overall top performance. As a result, it is evident that real-world suicidality detection using ML is invaluable. ML algorithms can analyze social media posts, text messages, and behavioral patterns to provide insights and cautions. This approach can provide a proper platform so that suicidality and mental illness do not threaten the fabric of our society anymore
ER  -