Walter J. Hernandez presented DLT-Corpus at KDD 2026 in Jeju, sharing a large-scale open text collection designed for research on distributed ledger technology.

‍

DLT-Corpus brings together 2.98 billion tokens from 22.12 million documents across scientific literature, patents and social media. The project also releases LedgerBERT, a domain-adapted language model that improves on BERT-base by 23 percent on a DLT-specific named-entity-recognition task.

‍

The work was developed with Peter Devine, Nikhil Vadgama, Paolo Tasca and Jiahua Xu, with support from Ripple’s University Blockchain Research Initiative.

‍

The open corpus and models give the wider research community new tools for studying how blockchain technology, markets and scientific knowledge develop over time.

‍