<![CDATA[The study aimed to evaluate sentiment classification models using toxicity scores and to conduct Social Network Analysis (SNA) to understand network dynamics. The research used CRISP-DM methodology to comprehensively analyze sentiment classification models and toxicity scores. It utilized various machine learning algorithms, including Decision Tree (DT), Support Vector Machine (SVM), and Naive Bayes Classifier (NBC), with Synthetic Minority Over-sampling Technique (SMOTE) to address class imbalance. In addition, Social Network Analysis (SNA) was conducted to examine network properties and dynamics. The findings revealed varying toxicity scores, ranging from 0.12409 to 0.98808, across different categories, such as general toxicity, severe toxicity, identity attacks, insults, profanity, and threats. Evaluation of sentiment classification models indicated that the SVM model with SMOTE achieved the highest accuracy of 92.57% +/- 1.17% (micro average: 92.57%), followed by the NBC model with an accuracy of 78.24% +/- 1.30% (micro average: 78.24%), and the DT model with an accuracy of 61.16% +/- 1.20% (micro average: 61.16%). Despite variations in model performance, the SVM model consistently demonstrated robust performance across various evaluation metrics. Furthermore, the SNA findings provided insights into network structural characteristics, including Average Degree, Average Weighted Degree, Diameter, Radius, and Average Path Length, facilitating a comprehensive understanding of network organization and behavior. These findings contribute to advancing the understanding of sentiment analysis models and network dynamics in digital environments.]]>
Copyrights © 2024
