Hybrid Deep Learning Architectural Framework for Analysis of Hateful Sentiment on Twitter (X)

Abstract

In recent years, the pervasive use of social media platforms, such as Twitter, has led to an exponential increase in the dissemination of information and opinions. However, this phenomenon has also given rise to the alarming prevalence of hateful sentiment, posing significant challenges for online communities and societal harmony. To address this issue, a Hybrid Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) Deep Learning Architectural Framework was designed specifically for the analysis of hateful sentiment in Twitter content. The framework combines the spatial feature extraction capabilities of CNNs with the sequential learning proficiency of LSTMs, aiming to capture both local patterns and long-term dependencies within the textual data. The framework achieves a comprehensive understanding of the sequential and contextual nature of language, thereby enhancing the accuracy of hateful sentiment analysis. To evaluate the effectiveness of the proposed framework, extensive experiments were conducted on two large datasets of labeled tweets containing hateful sentiments. The work was done in two tasks using different datasets and the results demonstrate superior performance in both tasks compared to existing state-of-the-art models. Task 1 generated a 0.82 macro F1-score and 0.92 weighted F1-score, and Task 2 generated a 0.82 macro F1-score and 0.90 weighted F1-score, showcasing the ability of the proposed architecture to discern subtle variations in hate speech, sarcasm, and disguised forms of harmful language. Beyond sentiment analysis, the implications of this research extend to the development of robust tools for fostering a safer online environment.