Abstract:

Abstract In line with technological advances, the increasing number of internet users caused more and more social media users, one of which is Twitter. Twitter as a social media is generally used by its users as a place to express themselves, both positive and negative expressions. One example of such negative expressions is depression. According to WHO research, depression is a common mental disorder that can lead the sufferers to commit suicide. In an effort to suppress this issue, many studies seek to develop machine learning models that can identify depression, especially in social media such as Twitter. This research aimed to identify indication of depression of Twitter user in Indonesia using text mining. Tweets which portraying stress are marked by a psychologist and TF-IDF are used for feature extraction in Traditional Machine Learning (TML) and for Deep Learning (DL), word tokenizer is used.  The classification methods used Multinomial Naive Bayes (MNB) and Support Vector Machine (SVM) for TML and Convolutional Neural Network for DL. Over-sampling and under-sampling methods and combination of both are used to deal with TML. The experimental results confirm that MNB, SVM and CNN can be used to detect tweet that has indication of depression in Indonesia. In addition, the use of Deep Learning (CNN) outperforms Traditional Machine Learning (MNB and SVM) in terms of performance scores with a longer calculation time. The highest accuracy of the model reached 91.23%.

Keywords: