paper2kg

Generate Knowledge Graph for Paper

Extractor/Algorithm

• Ollie, Java code extracts relation triples which has been integrated into the project.

Notice: BERT-BiLSTM-CRF will be added.

Files

• paperAPI.py: api entry
• d3.html: visualization demo (You need to replace the 'links' value [line 8, d3.html].)
• toolkit
  • Ollie: triple extractor
  • pdf_parser: pdf2xml

Requirements (python packages)

• uvicorn
• fastapi
• pydantic
• nltk

Before using NLTK, you may need to add the following code to the top of readxml.py and text2kg.py

nltk.download('punkt')
nltk.download('averaged_perceptron_tagger')

Once the download is successful, these two lines of code must be removed.

PS: JAVA environment is required.

Run

No API: Configure readXML.py and run.

or

uvicorn paperAPI:app --reload --port 8000 --host 127.0.0.1

or

python paperAPI.py

• GET: http://127.0.0.1:8000/paper2kg?paperID=ELG.pdf&confidence=0.6&fine_grain=false

• POST: {"paperID": "ELG.pdf", "confidence": 0.1, "fine_grain": "False"}

• API docs: http://127.0.0.1:8000/docs

Preview

Word Classes/tags for paper2api

• NNP: proper noun, singular
• NNPS: proper noun, plural
• NN: noun, singular
• NNS: noun, plural