HateDetector: Multilingual technique for the analysis and detection of online hate speech in social networks

With the proliferation of social media platforms that provide anonymity, easy access, and the establishment of online communities and discussion, hate speech identification and monitoring has become a major concern for society, individuals and policymakers, which can be interpreted as hate speech. Many researchers attempted to detect hate speeches in multiple languages from social media, but the research was limited due to high complexity and minimum accuracy. A novel 'HateDetector: Multilingual Hate Speech Detection Technique' has been proposed to overcome these issues. In this technique, Bidirectional Encoder Representations from Transformers (BERT) with Multi-Layer Perceptron (MLP) is developed to identify the nature of the tweets by performing the process of code conversion and similarity check that results in good vector values representing the tweet nature. Additionally, the exact sentiment or nature of a tweet, whether hate or non-hate, is identified using the Profanity Check Technique (PCT), composed of ReLu activation function with a logistic regression classifier that classifies the resultant vectors and its respective emoji to neutral or hate speech. This technique performs all analyses of a tweet. It also auto-detects and easily finds hate speech, even from poorly written and complex text. According to the experiment's findings, the proposed technique performed exceptionally well, with a classification accuracy of 97.9%. Our proposed technique was able to compete with other state-of-the-art models.