000			02938nam a22002657a 4500
003			KOHA
005			20250108101216.0
008			240927d2024 cy de||| |||| 00| 0 eng d
040			_aCY-NiCIU _beng _cCY-NiCIU _erda
041			_aeng
090			_aYL 3516 _bA33 2024
100	1		_aAhadi, Seyedeh Aridis
245	1	0	_aDETECTING INDICATORS OF MENTAL DISORDER IN SOCIAL MEDIA POSTS VIA HYBRID DEEP LEARNING AND NATURAL LANGUAGE PROCESSING / _cSEYEDEH ARIDIS AHADI ; SUPERVISOR, ASST. PROF. DR. KIAN JAZAYERI
264			_c2024
300			_a134 sheets ; _c30 cm _e+1 CD ROM
336			_2rdacontent _atext _btxt
337			_2rdamedia _aunmediated _bn
338			_2rdacarrier _avolume _bnc
502			_aThesis (MSc) - Cyprus International University. Institute of Graduate Studies and Research Information Technologies
520			_aIdentifying 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.
650		0	_aInformation Technologies _vDissertations, Academic
700	1		_aJazayeri, Kian _esupervısor
942			_2ddc _cTS
999			_c292975 _d292975