[Paper] Document Data Matching for Blockchain-Supported Real Estate

Source: arXiv - 2512.24457v1

Overview

The paper introduces a blockchain‑backed platform that automates the extraction, verification, and management of real‑estate documents. By chaining together OCR, NLP, and verifiable credentials (VCs), the authors aim to replace the error‑prone, paper‑heavy workflows that still dominate property transactions.

Key Contributions

• Unified OCR‑NLP pipeline trained on synthetic real‑estate documents, capable of handling diverse layouts (titles, deeds, contracts, etc.).
• Standardization layer that converts extracted fields into W3C‑compatible Verifiable Credentials, enabling interoperable data exchange.
• Automated data‑matching engine that cross‑checks multiple credentials to flag inconsistencies or potential fraud.
• Decentralized trust fabric built on a permissioned blockchain that stores credential hashes and audit trails, guaranteeing immutability and provenance.
• End‑to‑end prototype covering issuer, holder, and verifier roles, with a web UI that demonstrates real‑world transaction flows.

Methodology

1. Synthetic Dataset Generation – The team programmatically created thousands of mock property documents (varying fonts, languages, and scan qualities) to train the OCR model without exposing sensitive real data.
2. OCR + NLP Extraction – A lightweight OCR engine (Tesseract‑based) feeds raw text into a fine‑tuned BERT‑style NLP model that identifies key entities (owner name, parcel ID, sale price, etc.).
3. Credential Issuance – Extracted entities are mapped to a VC schema; the backend signs the credential with the issuer’s private key and records its hash on a Hyperledger Fabric network.
4. Data Matching & Verification – When a verifier receives multiple VCs (e.g., title deed + mortgage contract), a rule‑based matcher compares overlapping fields and raises alerts on mismatches.
5. User‑Facing Frontend – A React application implements the three roles:
   • Issuer: uploads scanned docs → triggers extraction → issues VCs.
   • Holder: stores VCs in a wallet (local encrypted storage).
   • Verifier: pulls VCs, runs the matcher, and displays a trust score.

Results & Findings

• The pipeline maintains >90 % accuracy even on low‑resolution scans, outperforming baseline OCR‑only approaches by ~5 pts.
• Credential issuance and blockchain anchoring add <0.5 s overhead, proving the solution is fast enough for interactive user experiences.
• The data‑matching engine successfully identified 87 % of injected inconsistencies in a controlled test, demonstrating its fraud‑detection potential.

Practical Implications

• Speed up closings – Real‑estate agents can cut document verification from days to seconds, accelerating cash flow and reducing escrow costs.
• Reduce fraud – Immutable credential hashes and automated cross‑checking make it harder to slip in forged deeds or altered mortgage terms.
• Interoperability – By adhering to open VC standards, the system can plug into existing property registries, title insurers, and fintech platforms without custom integrations.
• Developer‑friendly stack – The prototype uses widely adopted tools (Tesseract, Hugging Face Transformers, Hyperledger Fabric, React), lowering the barrier for teams to adopt or extend the solution.
• Scalable trust layer – Permissioned blockchain ensures that only authorized parties (government registries, banks) can write to the ledger, while anyone can verify the integrity of a credential.

Limitations & Future Work

• Synthetic‑data bias – Training on generated documents may not capture all quirks of legacy paper forms; a larger corpus of real scanned deeds is needed for robust generalization.
• Permissioned blockchain constraints – The current Hyperledger setup requires a consortium governance model; exploring public‑chain or layer‑2 alternatives could broaden adoption.
• Legal acceptance – While VCs are technically sound, regulatory frameworks for digital property titles vary by jurisdiction and will need alignment.
• Extending to multimodal inputs – Future versions could incorporate video walkthroughs or IoT sensor data (e.g., smart‑meter readings) to enrich the credential ecosystem.

Bottom line: By marrying OCR/NLP with verifiable credentials and blockchain, the authors deliver a practical blueprint for digitizing real‑estate paperwork—an area ripe for automation, transparency, and developer innovation.

Authors

• Henrique Lin
• Tiago Dias
• Miguel Correia

Paper Information

• arXiv ID: 2512.24457v1
• Categories: cs.CR, cs.DC
• Published: December 30, 2025
• PDF: Download PDF