Detection of depression and suicide in online social networks with deep learning and machine learning methods

In this thesis, deep learning and machine learning based text classifier models are built to detect suicidal thoughts of depressed people who post on online social networks (ASNs) and predict whether they will commit suicide. Three separate datasets, Suicide Ideation, Reddit Depression Suicidewatch and Suicide Detection, were used in the study. Suicide Ideation dataset is labeled as 'Potential suicide post' and 'Not suicide post', Reddit Depression Suicidewatch dataset is labeled as 'Depression' and 'SuicideWatch', and Suicide Detection dataset is labeled as 'Suicide' and 'Non-suicide'. These datasets facilitate the development and evaluation of deep learning and machine learning models for the binary classification task. In this study, Artificial Neural Networks (ANN), Deep Belief Networks (DBN), Convolutional Neural Networks (CNN), Long Short-Term Memory (LTSM), FastText, Naive Bayes (NB), Logistic Regression (LR), k-nearest neighbor (k-NN), Random Forest (RF) and Decision Tree (DT) methods were used to build suicidal ideation prediction models, and TF-IDF was applied to machine learning methods and FastText to deep learning models. In addition, Grid Search method was applied to all algorithms for hyperparameter optimization and the results were compared. As a result of the first comparisons of the thesis study, RF model with 94% in Suicide Ideation dataset, LR model with 73% in Reddit Depression Suicidewatch dataset and ANN and LR with 93% in Suicide Detection dataset were the most successful models. After applying the Grid Search metric, RF model with 94% in Suicide Ideation dataset, ANN model with 74% in Reddit Depression Suicidewatch dataset and ANN and LR with 93% in Suicide Ideation dataset were the most successful models.