Answers are grounded in this paper and cite their sources.
Demonstrates improved hate speech detection accuracy in code-mixed languages, suggesting better comprehension of context and sarcasm.
Hate speech detection remains challenging due to the complexity of language, sarcasm, and the increasing use of emojis and code-mixed text. Hence, a novel Selective Pictographic Filter and Conjoint Semantic Topic Comparator is proposed to improve the detection accuracy of sarcastic hate speech and contextual information in a code-mixed Hindi-Marathi-English dataset. In the pre-processing stage, existing approaches remove emojis. This misses essential cues for sarcasm. To address this, a Selective Trimming Pictographic Filter is proposed that uses an Emoji Enchant Library Screening strategy to retain unique, relevant emojis, preserving the semantic value for detecting polite, sarcastic hate speech. Another challenge arises from the intricacies of code-mixed languages, where informal transliteration and non-standard spelling complicate vocabulary extraction. Distractive Attention FastText-based ADM (Automatic Decision Making) is introduced, which extracts robust vocabulary and eliminates informal transliteration, enabling better comprehension of context in mixed-language posts. Moreover, existing transformer-based models are constrained by a fixed maximum sequence length, which limits their ability to fully capture long-range contextual information in indirect hate speech detection. To tackle this, the Correlated Hierarchical Dirichlet Process (CHDP) comparator further analyses the semantics of both text and emojis, identifying latent topics and their relations. Finally, a Flexible Differential Neuro Classifier is developed, which extends the context window dynamically using a Differentiable Neural Computer Network, thus improving the ability to capture complex contextual signals, leading to more accurate detection of sarcastic hate speech. As a result, the proposed model outperforms existing methods, achieving higher accuracy, precision, recall, and AUC values.
No takes yet. Share an insight, caveat, or question.
Chavan et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: