A Sentiment Analysis Approach for Abusive Content Detection using Improved Dataset

Abstract

The rapid growth of information and communications technologies has led to the generation of enormous amount of information daily. Consequently, there has been an increased interest in effective data processing. The nature of the information is varied and can in some cases include users’ emotions and opinions. Faced with this situation, the need of proposing social media monitoring and content filtering is a major asset for the community. The specific case of abusive content is becoming more relevant these recent years leading to the proposition of many models which unfortunately suffer from the unbalanced nature of the dataset used. We propose a sentiment analysis approach which classifies social media posts according to three categories: hate, abusive and neutral. The approach is based on a constructed dataset which reduces unbalancing and improves classification results.