Blockchain Hate Logger: Immutable Logging of Detected Hate Speech Using DeHateBERT

Overview & Synopsis

The rise of hate speech on social media platforms presents significant challenges for content moderation, transparency, and user trust. Although hate speech classification has been widely explored using models such as DeHateBERT, existing moderation systems often lack accountability, and flagged content can be altered or removed without trace. In this paper, we present Blockchain Hate Logger a system that detects and immutably logs hate speech using a transformer-based text classifier (DeHate- BERT) and Ethereum smart contracts. We use the HateXplain dataset to classify 25,000 tweets into normal, offensive, or hate categories. For each flagged tweet, we compute a SHA-256 hash and store it along with the predicted label, model confidence, and timestamp on a smart contract deployed via Ganache, a local Ethereum blockchain. This ensures tamper-proof, verifiable moderation logs. We analyze over 1,000 transactions, estimate gas costs, and present trends in label distribution, confidence scores, and time-based logging. Our work builds on prior efforts in hate speech detection, fairness, and explainability, and demonstrates the feasibility of integrating NLP and blockchain for auditable AI content moderation. The full implementation is available at: https://github.com/mnusrat786/realtime-hate-logger.