Hate Speech Detection in Hinglish Text using Deep Learning

Abstract

Humanity is fortunate to have the Internet, but it has also becoming more exploited. Social networking sites like Instagram and Twitter are where users most frequently express their ideas. It may annoy readers when users use harsh or inflammatory language. Code-switching is usually semantically difficult in linguistically diverse, low-resource languages, and there aren't many sophisticated methods for reliably detecting hate speech in real-world data.The goal of this study is to evaluate and compare the effectiveness of a number of deep learning algorithms created to find instances of hate speech on popular social media sites that use Hinglish (an English-Hindi code-mix) language. This article's goal is to look at a number of deep learning algorithms for identifying hate speech on popular social media platforms in English, Hindi, and Hinglish. In order to identify hate speech from tweets and comments in Hinglish, English, and Hindi tweets, we implement and analyse various deep learning approaches as well as a number of word-embedding techniques (Glove, Fasttext, Word2vec) using a consolidated dataset of about 21800 occurrences. In this paper, we applied and evaluated several deep learning algorithms along with different embedding techniques on a amalgamated dataset of 21748 instances, for speech recognition from comments, tweets, etc. CNN-Bi-LSTM with Fasttext word embedding technique provides the best results. It yield accuracy(0.72), precision(0.69), recall(0.69), F1-score(0.72) and ROC- AUC(0.76).