Sentiment Analysis Using Machine Learning Methods on News Texts

The categorization of intricate emotional phenomena in news texts permits the assessment of the impact of such news on society. In this context, it is evident that the existing body of research on emotion analysis in Turkish texts is insufficient. The objective of this study is to identify the emotions of "fear," "happiness," "anger," "sadness," and "surprise" in Turkish news texts using machine learning methods. In consideration of the Turkish language structure, the adjectives and nouns were parsed on a specially prepared corpus and dictionary groups specific to this study were created with the use of the TF-IDF and Double Normalization methods. Subsequently, these lexicon groups were employed for the classification of sentiment in news texts using a range of machine learning techniques, including Multilayer Perceptron (MLP), Support Vector Machines (SVM), Random Forest (RF), LightGBM, XGBoost, GPT-2, BERT, DistilBERT, and XLM ROBERTa. The evaluation metrics employed were F1-score, accuracy, recall, and precision. In the training phase of the model, the dictionary group created with Double Normalization, the GPT-2 model, the Grid Search optimization method, and cross-validation were employed in order to achieve the highest F1-score of 0.99 and accuracy of 0.99. It is anticipated that the methods and dataset utilized in this study will contribute to the existing literature on sentiment classification in Turkish texts and provide insights that will inform future similar studies.