Sentiment Analysis onWomen’s Rights Discussions in Turkish Social Media Using Machine Learning Methods

Gender-based legal protections and women’s rights frameworks have been a major topic of public debate in T¨urkiye, with intense discussions occurring on social media. Understanding the sentiment behind such online discussions can help capture public perceptions and their evolution over time. Therefore, this study aims to detect sentiment in Turkish socio-political posts using machine learning techniques. A total of 15,943 posts from 2010 to 2024 were collected and preprocessed, resulting in a clean dataset of 13,008 posts. Within the scope of the study, sentiments were labeled as positive, negative, or neutral using both BERT-based models, and classification was performed using basic machine learning algorithms. First, two pre-trained BERT-based models were utilized for sentiment analysis, with the selected model achieving 71.9% accuracy on a manually labeled test set. Then, further analysis was conducted using traditional machine learning algorithms, where Random Forest achieved the best performance with a 70.5% accuracy on the manually labeled test set. To analyze the perception of the human rights treaties in society, time-based sentiment shifts, especially around key political events, were visualized. The visualizations show that negative posts received more interaction than positive posts, and the number of negative posts was considerably higher than neutral and positive posts during the important dates for the issue. This study demonstrates the value of combining social media data with machine learning to interpret public opinion on significant societal debates.