A Multiclass Approach to Identify Misogynistic Bangla Text from Social Media

Misogyny is defined as hostility, dominance, harassment, intimidation, and violence against women. Social media networks are rapidly developing. As a result, misogyny is becoming increasingly fashionable. This paper presents a solution to the problem of automatically classifying misogynistic Bengali messages on social media platforms. Several procedures have been taken to preprocess the data. Tf-Idf and Word Embedding with BERT are implemented so that computers can read text. Several machine learning-based models including Random Forest, SVM Polynomial kernel, SVM Signal kernel, and Adaptive Boosting have been tested to build a multi-class analyzer that identifies Misogynistic text in Bengali on social media. The models have been evaluated with the performance measures such as accuracy, precision, recall, and F1 scores. With 62.61 percent accuracy, the SVM sigmoid kernel outperforms all other models. As per our knowledge, no earlier research has been conducted on the classification of Bengali misogynistic text; consequently, the current research is the most comprehensive to date.