The internet and social media facilitate widespread idea sharing but also contribute to cyber-crimes and harmful behaviors, notably the dissemination of abusive and hateful speech, which poses a significant threat to societal cohesion. Hence, prompt and accurate detection of such harmful content is crucial. To address this issue, our study introduces a fully automated end-to-end model for hate speech detection and classification using Natural Language Processing and Deep Learning techniques. The proposed architecture comprising embedding, Convolutional, bidirectional Recurrent Neural Network, and bidirectional Long Short Term Memory layers, achieved the highest accuracy of 98.5%. Additionally, we employ explainable AI techniques, such as SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME), to gain insights into the performance of the proposed framework. This comprehensive approach meets the pressing demand for swift and precise detection and categorization of harmful online content.

中文翻译：

---

迈向更安全的在线社区：用于仇恨言论检测和分类的深度学习和可解释的人工智能

互联网和社交媒体促进了广泛的思想共享，但也助长了网络犯罪和有害行为，特别是辱骂和仇恨言论的传播，这对社会凝聚力构成了重大威胁。因此，及时、准确地检测此类有害内容至关重要。为了解决这个问题，我们的研究引入了一种使用自然语言处理和深度学习技术进行仇恨言论检测和分类的全自动端到端模型。所提出的架构包括嵌入、卷积、双向循环神经网络和双向长短期记忆层，实现了 98.5% 的最高准确率。此外，我们还采用可解释的人工智能技术，例如 SHapley Additive exPlanations (SHAP) 和 Local Interpretable Model-agnostic Explanations (LIME)，来深入了解所提出的框架的性能。这种综合方法满足了快速、准确地检测和分类有害在线内容的迫切需求。