Human-computer interaction will be revolutionized by the emotion-aware artificial intelligence systems, but it has severe privacy-preserving deployment challenges. The proposed work introduces a scalable solution for offline NLP emotion identification with identity verification via blockchain, bridging a critical gap between high-accuracy affective computing and user privacy. The system was tested on a diverse dataset of 21, 000 English sentences. By utilizing “j-hartmann/emotion-english-distilroberta-base” as a transformer, we were able to achieve state-of-the-art performance in seven-class emotion detection with a mean accuracy 94. 2\% 1. 8 and mean confidence recorded above 0. 88 with joy and anger classes with low computational cost, making it edge-friendly. Benchmark evaluation on the GoEmotions dataset achieved a competitive macro F1-score of 65. 9% compared to existing state-of-the-art models, while it maintains the edge-compatibility throughput of 1200 sentences/second. In our examination, we found that the model is robust and capable of handling actual linguistic diversities across various fields. Anchoring hashed emotion profiles to decentralized identities by the proposed system architecture permits (1) tamper-proof storage of emotional contexts on Ethereum-based smart contracts, (2) resistance to Sybil attacks by behavioral-emotional fingerprinting, and (3) compliance with GDPR through personalization. On benchmark systems, the lightweight “DistilRoBERTa” version can sustain 1, 200 sentences/second on consumer systems, which proves that it is feasible to deploy at scale. To our knowledge, it is one of the first integrated frameworks that pairs offline emotion recognition with decentralized identity (DIDs) verification, combined with a strong privacy-preserving layer. We believe the study is capable of enabling secure affective computing employing blockchain architecture and making it impactful for many applications, including mental health tracking, customized learning, and fraud-free conversational AI.
Hasan et al. (Thu,) studied this question.