Extracting Entities and Relations in Portuguese Documents
summary
A pipeline extracts names, IDs and powers from Portuguese Letters of Attorney and Articles of Incorporation using Google Vision OCR, a POS tagger (Pavel), CRF-based NER and a CNN relation model with word embeddings. The system links who grants which power across documents and achieves about 80% F1 on Articles of Incorporation while struggling with OCR errors on low-quality scans and sparse training data; heavy annotation effort proves necessary for reliable results.
AI-written summary · about
chapters (6)
related talks
48mBuilding practical natural language processing systems
44mExtracting Structured Data from Legal Documents
58mRelation Extraction from Online PH News Source
01h 30mRelation extraction from texts
49mNamed-Entity Recognition and Name Resolution Using Similarity Queries
42mUsing Deep Learning to Understand Documents
15mFrom Image to Graph: Leveraging Tesseract OCR Engine for Document Chunking and GraphRAG
29mdeepdoctection - An open source package for document intelligence