Abstract
The world is witnessing revolutionary evolution of internet and with the advent of social media; users are empowered to easily post contents on the web at any time and from any place in the form of opinions, comments, and feelings. Manual approaches of detecting and analyzing such huge amount of posts are not feasible and there is a need for automated methods and techniques to discover the knowledge and patterns of the text content without human involvement. Text mining refers to the process of extracting interesting and significant patterns or knowledge from text documents. YouTube is known for its free provision of video sharing service. The content of YouTube videos may sometimes comprise of images or sequence(s) of images with unwanted material, such as aggression, which is the reason of emergence of many social problems, particularly among children such as demonstration of aggressive behavior and bullying at home, school and public places. The research work reports performance of machine learning classifiers that were applied on video transcripts of YouTube videos to detect aggression. The dataset constructed for the purpose of research work, consists of English video scenes transcripts that were collected from the web and were annotated manually as violent and non-violent. Various experiments were performed on the dataset using different machine learning (ML) classifiers with different text preprocessing settings in RapidMiner and Python environments and thus predictive classifier models were constructed and tested. In RapidMiner environment, the SVM classifier model outperformed the other classifiers achieving highest accuracy of 79% after preprocessing step of removal of stop words. In Python programming environment, NB classifier outperformed the other classifiers in majority of experiments with different preprocessing settings, achieving highest accuracy of 82.5%, when stemming was performed in preprocessing stage along with other preprocessing steps. The automatic process of aggression detection in video scenes can be used by concerned authorities to enforce their cultural priorities.