Explainable Ai for Offensive Content Detection and Analysis on Social Media

The widespread use of online platforms has led to an increase in inappropriate content, posing significant challenges to digital safety. This paper presents the development of a chatbot designed for offensive content analysis in online conversations. Utilizing advanced machine learning techniques, the chatbot employs a Support Vector Classifier (SVC) combined with Word2Vec for detecting offensive language. Additionally, a LIME model is introduced to enhance interpretability by analyzing the contribution of different words toward the offensiveness score. As a second phase, from the detected offensive content a Logistic Regression-based multi-label model is used to assess toxicity levels. Trained on a dataset of Twitter messages labeled for offensive and non-offensive content, the system effectively identifies and classifies harmful language. With high accuracy and computational efficiency, this solution enhances content moderation efforts and promotes a safer online environment.