An Efficient Hate Speech Detection Method Using an Optimized LSTM Based on a Hybrid Efficient Document Representation Method

K. Senthil Kumar, S. Sathiyabama, D. Ayyamuthukumar

Abstract: As the popularity of social media platforms surges, the problem of hate speech has turned out to be a serious issue, endangering the safety of the world online, social cohesion, and the well-being of individuals. Hate speech is difficult to detect on such platforms as Twitter because the data is very short, informal, and noisy, usually containing slang and abbreviations, emojis, and vague phrases. Simple neural networks and traditional machine learning techniques are not particularly effective in learning highly complex semantic and contextual patterns needed to reach high-quality classification. An alternative to these difficulties is that this study suggests a hybrid model of Elman Recurrent Neural Network (ERNN) that is optimized by the Local search algorithm (LSA) and Elephant Herding Optimization (EHO). The model uses the best document representation methods, such as Doc2Vec, TF-IDF, BERT, and a hybrid of TF-IDF + BERT representation, to identify rich semantic, syntactic, and contextual features of tweets. Experiments on benchmark Twitter datasets have shown that, compared to ERNN as a baseline and other ERNN variants that are metaheuristically optimized, the proposed ERNN-LSA-EHO framework has better accuracy, precision, recall, and F1-score. The hybrid optimization and sophisticated feature representations contribute to effective convergence, robust learning, and the ability to detect very subtle and complex patterns of hate speech. The results of the research demonstrate the possibility of the given method to monitor social media in real time, moderate content, and perform sentiment analysis.

Keywords: Hate Speech Detection, Elman Recurrent Neural Network (ERNN), Local Search Algorithm (LSA), Elephant Herding Optimization (EHO), Document representation, Social Media Analysis

How to Cite?: K. Senthil Kumar, S. Sathiyabama, D. Ayyamuthukumar, "An Efficient Hate Speech Detection Method Using an Optimized LSTM Based on a Hybrid Efficient Document Representation Method", Volume 14 Issue 12, December 2025, International Journal of Science and Research (IJSR), Pages: 1185-1197, https://www.ijsr.net/getabstract.php?paperid=SR251215115834, DOI: https://dx.doi.org/10.21275/SR251215115834