Antisocial Behaviour Analyses Using Deep Learning

摘要

Online antisocial behaviour is a social problem and a public health threat. It is one of the ten personality disorders and entails a permeating pattern of violation of the rights of others, and disregard for safety. It prevails online in the form of aggression, irritability, lack of remorse, impulsivity, and unlawful behaviour. The paper introduces a deep learning-based approach to automatically detect and classify antisocial behaviour (ASB) from online platforms and to generate insights into its various widespread forms. Once detected, appropriate measures can be taken to eradicate such behaviour online and to encourage participation. The data for this paper was collected over a period of four months from the popular online social media platform Twitter by using pre-defined phrases linked to antisocial behaviour. Widely used machine learning classifiers: SVM, Decision tree, Random Forest, Linear regression, and deep learning architecture (CNN) were experimented with. CNN was implemented with both GloVe and Word2Vec embeddings and outperformed all the traditional machine algorithms used in the study. Standard performance metrics such as accuracy, recall, precision, and f-measures were used to evaluate classifiers and the CNN-GloVe combination (with 300 dimensions) produced the highest classification performance achieving 98.42% accuracy. Visually enhanced interpretation of the results is presented to demonstrate the inner workings of the classification process.