The proliferation of digital assets within global e-commerce ecosystems has intensified interest in understanding the determinants of consumer adoption behavior. This study presents an empirical framework employing Bidirectional Encoder Representations from Transformers (BERT), operationalized through a domain-adapted variant designated CryptoBERT, for the sentiment classification of user-generated social media content, and examines the predictive utility of BERT-derived sentiment scores for cryptocurrency adoption intention in e-commerce contexts. Drawing on a corpus of 148,130 posts harvested from Twitter/X and Reddit spanning January 2024 through December 2025—encompassing the Bitcoin spot ETF activation period, the April 2024 halving, the Ethereum ETF launch (July 2024), and the 2025 bull market cycle—and grounded in the Technology Acceptance Model (TAM) and the Unified Theory of Acceptance and Use of Technology (UTAUT2), CryptoBERT is fine-tuned on a manually annotated gold-standard corpus of 18,500 posts using MATLAB's Text Analytics Toolbox for preprocessing, feature extraction and performance benchmarking. The model attains a macro-averaged F1-score of 91.7%, exceeding BiLSTM (84.3%), TextCNN (82.1%), and lexicon-based VADER (73.6%) baselines by statistically significant margins. Sentiment scores are subsequently incorporated as independent variables in a binary logistic regression adoption prediction model, yielding an AUC-ROC of 0.893 and Nagelkerke R² of 0.487. Results indicate that positive sentiment coheres around financial inclusivity, transaction efficiency, and merchant participation, while negative sentiment is predominantly driven by price volatility, regulatory opacity, and exchange failures. The framework provides a scalable, data-driven instrument for merchants, policymakers, and fintech innovators seeking to monitor and shape cryptocurrency adoption trajectories.