Beyond Benchmarks: Exposing the Hidden Crisis in Bangla Hate Speech Detection

• URL: http://arxiv.org/abs/2607.11597v1
• Date: Mon, 13 Jul 2026 14:19:16 GMT
• Title: Beyond Benchmarks: Exposing the Hidden Crisis in Bangla Hate Speech Detection
• Abstract summary: Hate speech spread on social media platforms poses a major concern for online safety and ethical moderation.<n>This study aimed to expose the crisis of Bangla HS detection systems by diagnosing how and why benchmark-trained models fail to identify implicit, context-dependent HS.<n>Emoji-aware preprocessing improved implicit HS detection by up to 12%, whereas emoji removal caused a notable decline in performance.
• Score: 3.6609212521250227
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: The spread of hate speech (HS) across different social media platforms (SMPs) poses a major concern for online safety and ethical moderation. Automatic detection of HS remains a challenging task, especially in under-resourced languages like Bangla, due to cultural context, implicit expressions, and informal linguistic patterns. This study aimed to expose the crisis of Bangla HS detection systems by diagnosing how and why benchmark-trained models fail to identify implicit, context-dependent HS. Six architectures (FastText + CNN, FastText + LSTM, FastText + BiLSTM, BanglaBERT, BanglaBERT + CNN, and BanglaBERT + BiLSTM) were trained on benchmark datasets (about 75,000 posts) and a merged multi-source dataset (about 120,000 posts), then externally validated on an annotated dataset (about 200 posts) collected from Facebook, Twitter, and YouTube, labeled as HS and non-HS, where HS was further categorized as explicit and implicit. BanglaBERT achieved an F1-score of 91.4% on benchmark datasets but declined to 75.3% on the external set and 63.4% for implicit HS involving sarcasm and emojis. The accuracy of FastText + CNN dropped from 78.0% to 51.2% under similar conditions. Emoji-aware preprocessing improved implicit HS detection by up to 12%, whereas emoji removal caused a notable decline in performance (F1: 0.75 to 0.63). Frequent misclassifications in politically charged or satirical comments revealed over-policing risks. This study not only exposes the generalization crisis due to implicit, culturally embedded, and emoji-laden expressions but also underscores the need for developing adaptive, emoji-aware, and culturally grounded frameworks that ensure ethical moderation while preserving freedom of expression. Findings of this study provide insights for researchers, SMPs, and policymakers to design more context-sensitive HS detection systems for low-resource languages.

Related papers

• DIA-HARM: Dialectal Disparities in Harmful Content Detection Across 50 English Dialects [6.107850985025956]
  We present DIA-HARM, the first benchmark for evaluating disinformation detection across 50 English dialects.<n>Human-written dialectal content degrades detection by 1.4-3.6% F1, while AI-generated content remains stable.<n>We release the DIA-HARM framework, D3 corpus, and evaluation tools.
  arXiv  Detail & Related papers  (2026-04-07T01:43:48Z)
• Lingua-SafetyBench: A Benchmark for Safety Evaluation of Multilingual Vision-Language Models [54.10540442330978]
  Existing benchmarks are typically multilingual but text-only, or multimodal but monolingual.<n>Recent multilingual red-teaming efforts render harmful prompts into images, yet rely heavily on typography-style visuals.<n>We introduce a benchmark of 100,440 harmful image-text pairs across 10 languages, explicitly partitioned into image-dominant and text-dominant subsets.
  arXiv  Detail & Related papers  (2026-01-30T09:18:13Z)
• Improving Code-Switching Speech Recognition with TTS Data Augmentation [58.34842693152991]
  This paper explores multilingual text-to-speech (TTS) models as an effective data augmentation technique to address this shortage.<n>We fine-tune the multilingual CosyVoice2 TTS model on the SEAME dataset to generate synthetic conversational Chinese-English code-switching speech.
  arXiv  Detail & Related papers  (2026-01-02T10:11:51Z)
• Cyberbullying Detection in Hinglish Text Using MURIL and Explainable AI [0.356008609689971]
  The rise of code-mixed Hindi-English (Hinglish) communication on digital platforms poses challenges for existing cyberbullying detection systems.<n>This paper presents a framework for cyberbullying detection in Hinglish text using the Multilingual Representations for Indian languages (MURIL) architecture.
  arXiv  Detail & Related papers  (2025-06-19T06:46:42Z)
• POSTER: A Multi-Signal Model for Detecting Evasive Smishing [2.7039386580759666]
  We present a multi-channel smishing detection model that combines country-specific semantic tagging, structural pattern tagging, character-level stylistic cues, and contextual phrase embeddings.<n>We curated and relabeled over 84,000 messages across five datasets, including 24,086 smishing samples.<n>Our unified architecture achieves 97.89% accuracy, an F1 score of 0.963, and an AUC of 99.73%, outperforming single-stream models by capturing diverse linguistic and structural cues.
  arXiv  Detail & Related papers  (2025-05-23T12:45:34Z)
• Text Injection for Neural Contextual Biasing [57.589903308622745]
  This work proposes contextual text injection (CTI) to enhance contextual ASR. CTI with 100 billion text sentences can achieve up to 43.3% relative WER reduction from a strong neural biasing model.
  arXiv  Detail & Related papers  (2024-06-05T04:20:17Z)
• CBBQ: A Chinese Bias Benchmark Dataset Curated with Human-AI Collaboration for Large Language Models [52.25049362267279]
  We present a Chinese Bias Benchmark dataset that consists of over 100K questions jointly constructed by human experts and generative language models. The testing instances in the dataset are automatically derived from 3K+ high-quality templates manually authored with stringent quality control. Extensive experiments demonstrate the effectiveness of the dataset in detecting model bias, with all 10 publicly available Chinese large language models exhibiting strong bias in certain categories.
  arXiv  Detail & Related papers  (2023-06-28T14:14:44Z)
• Overview of Abusive and Threatening Language Detection in Urdu at FIRE 2021 [50.591267188664666]
  We present two shared tasks of abusive and threatening language detection for the Urdu language. We present two manually annotated datasets containing tweets labelled as (i) Abusive and Non-Abusive, and (ii) Threatening and Non-Threatening. For both subtasks, m-Bert based transformer model showed the best performance.
  arXiv  Detail & Related papers  (2022-07-14T07:38:13Z)
• BD-SHS: A Benchmark Dataset for Learning to Detect Online Bangla Hate Speech in Different Social Contexts [1.5483942282713241]
  This paper introduces a large manually labeled dataset that includes Hate Speech in different social contexts. The dataset includes more than 50,200 offensive comments crawled from online social networking sites. In experiments, we found that a word embedding trained exclusively using 1.47 million comments consistently resulted in better modeling of HS detection.
  arXiv  Detail & Related papers  (2022-06-01T10:10:15Z)
• COLD: A Benchmark for Chinese Offensive Language Detection [54.60909500459201]
  We use COLDataset, a Chinese offensive language dataset with 37k annotated sentences. We also propose textscCOLDetector to study output offensiveness of popular Chinese language models. Our resources and analyses are intended to help detoxify the Chinese online communities and evaluate the safety performance of generative language models.
  arXiv  Detail & Related papers  (2022-01-16T11:47:23Z)
• HS-BAN: A Benchmark Dataset of Social Media Comments for Hate Speech Detection in Bangla [2.055204980188575]
  In this paper, we present HS-BAN, a binary class hate speech dataset in Bangla language consisting of more than 50,000 labeled comments. We explore traditional linguistic features and neural network-based methods to develop a benchmark system for hate speech detection. Our benchmark shows that a Bi-LSTM model on top of the FastText informal word embedding achieved 86.78% F1-score.
  arXiv  Detail & Related papers  (2021-12-03T13:35:18Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.