NER-EDI

Named Entity Recognition for EDI, AI & data integration applied to EDI.

Definition

NER (Named Entity Recognition) applied to EDI consists of extracting structured entities from raw text: company name, IBAN, BIC, GTIN, VAT code, amount. Classical models: spaCy custom, fine-tuned BERT, HuggingFace transformers. Essential for OCR → JSON EDI pipelines. Typical precision: 90% on clear invoices, 75% on order emails.

Origin

Classical NLP discipline since MUC (Message Understanding Conference) 1996. Modernised with BERT 2018, and LLMs since 2023.

Example in context

spaCy custom NER with IBAN, BIC, AMOUNT, COMPANY entities trained on 5000 FR/EN emails, F1 = 0.89.

Related terms

• Document AI — service encompassing NER.