Analysis of Offensive Data over Multi-Source Social Media Environment Using Modified Random Forest Algorithm

Abstract

The widespread usage of social media platforms has resulted in an increasing volume of offensive content, posing significant challenges to maintaining a safe and respectful online environment. This research presents an analysis of offensive data over the social media environment using a modified Random Forest algorithm. The proposed modification to the traditional Random Forest algorithm incorporates a Weighted class Random Forest (WRF) to enhance model diversity and robustness. An algorithm utilizes weighted classes during training to address the inherent class imbalance in offensive data. By assigning higher weights to offensive content, the model prioritizes accurately identifying offensive posts, comments, and messages. This paper used the Twitter and Reddit dataset of multi-source social media content, labeled for offensive and non-offensive content, to train and validate the modified Random Forest model. Our proposed model is compared with Decision Tree (DT), Extreme-Gradient Boosting (XGBoost), Multi-Layer Perceptron (MLP), K-Nearest Neighbors (KNN), and Traditional Random Forest (RF) algorithms in machine learning. A number of performance metrics are used to assess the model's effectiveness in dealing with offensive data, including accuracy, recall, precision, specificity, and the F1-score. The results demonstrate that the modified Random Forest algorithm outperforms better than other machine learning algorithms. Moreover, the model shows improved resilience to variations in offensive language and context, making it more suitable for real-world applications.