Confronting Hate Speech in SMART Environments: An Approach that Uses Ensemble Learning and LSTM
Priyanka Kaushik, Rohit Rohilla, P. Walia, S. Vishnu Shankar, Manas Manglik Kaushik, Tapas K. Das Gupta
Abstract
In the current digital age of the Internet and online virtualization, it has established different kinds of online media, such as social media: two important and most pervasive ways of communicating. However, it becomes a little hard for them to deal with the responsibility, i.e., to regulate and filter the toxic or harmful content across the platform, specifically in filtering out the offensive comments posted by the users. In this work, we are particularly interested in the main mission of hate comment detection. Our major objective is to flag those comments which are offensive, harmful, and abusive in an online conversation. This paper proposes an original ensembling based learning scheme, taking the power of LSTM neural networks, to solve this crucial problem. LSTMs, specially designed types of RNNs, have shown to be effective in sequence modeling and also demonstrated some promising results in problems related to sequences. Our methodology is based on the existence of certain LSTM models of small size, which are trained on an independent fraction of all completed and pre-processed data. This approach allows the ensemble to capture the diversity in patterns; hence, it contributes to the large robustness of hate comment detection. In an ever-globalized and digitized world, the advent of dependable methodologies for hate comment detection becomes cardinal in acquiring safe and inclusive online environments. Our research is a promising step towards reducing online toxicity in order to have healthier and more constructive digital conversations.