Welcome to the MTE wiki!

The MTE system has three components:

1. The MTE database (stored as an SQLite database following the MTE Database Schema
2. The MTE ingestion pipeline (which populates the database)
3. The MTE user interface (website) - used internally at JPL; not part of this repository

Ingestion Pipeline

• The pipeline takes in a PDF file and applies these steps
  • Convert PDF to text (using Tika)
  • Obtain document information such as title, authors, etc. (lookup via ADS API and fall back to extraction from the text content using Grobid)
  • Extract (recognize) named entities, including Targets, Elements, Minerals, and Properties
    • Custom CoreNLP NER model
    • Custom Python CRFSuite model
  • Extract relations between entities (e.g., "contains") (using jSRE)
• To generate an MTE database: see detailed instructions
• MTE Wishlist - ideas for future improvements

Howto

• Update CoreNLP NER and restart the CoreNLP server