ASocTweetPred: Mining and Prediction of Anti-social and Abusive Tweets for Anti-social Behavior Detection Using Selective Preferential Learning

Abstract

In the present times, owing to a large number of social interactions via Online Social Networking platforms. Because the content is not moderated on the World Wide Web, especially on the Social Web, due to lack of moderation, there are many derogatory, Anti-Social, and racist comments. Specifically, Twitter, YouTube, Instagram, Facebook comments, and similar social network platforms. In this paper, ASocTweetPred for Mining and Prediction of Anti-Social and Abusive Tweets for Anti-Social Behavior Detection has been put forth by encompassing Ontologies of different phases, Knowledge Graph, and the classification of the dataset using the LSTM to generate and yield matching ontology snippets by computing the Semantic Similarity using the Bose Einstein’s Index and the APMI measure. Subsequently, the matched ontology snippets and the formulated Knowledge Graph are used in order to extract features by preferential selection using Shannon’s Entropy and Horn’s Index with differential step deviation measures. A triadic hybrid model of the LSTM, Bagging with Random Forest and SVC, and XGBoost classifiers are made use for Anti-Social Tweets prediction. An overall Precision of 97.93%, Recall of 98.09%, Accuracy of 98.01%, F-Measure of 98.00%, and with the lowest FNR of 0.02 has been achieved by the proposed ASocTweetPred framework.