Turkish tweet sentiment analysis with classical machine learning algorithms and transformer model

With the onset of the technological age, many applications have become popular today. One of these is Twitter (Newly known as X). Thanks to this platform, many users can share their personal thoughts and ideas with text, pictures or videos. These shared data attract the attention of many people. Scientists, academicians or other professionals who want to make personal inferences from these data do their work in this direction. For example, a police officer who wants to create a criminal profile can use tweets to do this, or an advertising agency can create personalized advertisements by looking at what a person writes or the pictures he/she shares, and thus product sellers can sell their products quickly. In this thesis, a sentiment analysis study was conducted in which we received a high accuracy rate. In this study, it was aimed to classify emotions using a ready-made Turkish Tweet data. This data set includes 4000 data which is separated by 5 different labels. Pre-processing was applied on this raw dataset. The pre-processed dataset was divided into training and testing. Performances were measured with classical machine learning models. These performances were measured in terms of accuracy, precision, recall, and F1 score as well as macro and weighted averages. Additionally, the confusion matrix calculated for each algorithm is given. The Stack algorithm achieved the highest accuracy rate of 96.88%. Also in this study, a pre-trained Transformer model, which is in the field of deep learning, was also used. The data set is divided into training, validation and testing. Likewise, its performance was measured in terms of accuracy, precision, recall and F1 score, and the confusion matrix calculated for each algorithm is given. 93% accuracy rate was achieved with this model.