Multiclass Emotion Analysis Of Public Perception Of Mpox Disease In X (Twitter) Data

This study aims to conduct sentiment analysis based on opinions expressed about monkeypox (mpox) in tweets shared on the X platform. To this end, 51,894 tweets containing the keywords “mpox” and “monkeypox” were collected from the X platform between January and June 2025. During the data collecting procedure, a keyword-based scanning strategy and API-based data retrieval techniques were used to retrieve English tweets. The raw data underwent extensive preprocessing to guarantee data quality and enhance the model's classification performance. These actions included changing the content to lowercase, eliminating URLs, user tags, and retweet phrases, and converting emojis to their textual equivalents. After these processes, a consistent dataset of 47,464 tweets suitable for analysis was obtained. The collected tweets were labeled with seven different categories describing the emotions they contained: fear, surprise, joy, sadness, anger, disgust, and neutrality. In the labeling process, automatic pre-labeling was performed using a zero-shot classification method, followed by manual verification by three independent experts. Inter-rater agreement was measured using Fleiss Kappa, and a high reliability level of 0.78 was achieved. The classification process was performed using five different Transformer-based models. The XLNet model demonstrated the highest performance, achieving an average accuracy of 99.05%, an F1 score of 95.64%, and a weighted F1 score of 99.05%. In conclusion, this study examined the public's emotional perception of global health threats such as mpox through social media using multi-class sentiment analysis. These findings will contribute to a better understanding of risk perception in the context of public health communication.