yardstick17 / knowledge-graph-on-research-paper Goto Github PK

Link: here

PYTHONPATH=. python main.py -c config/build_research_paper_knowledge_graph.yaml

src/output/kg_plot.png

Example path: /home/kushwaha/Projects/knowledge-graph-on-research-paper/src/output/kg_plot.png

Scalable code structure to run number of experiments without loosing their config. Reproduce them by making a new config to reproduce later. Add the steps from in the config:steps to execute them in the linear way.

data:
  research_paper_path: data/text_paper/aging.yaml

output:
  output_kg_plot_path: output/kg_plot.png

steps:
  - read-research-paper
  - document-relation-extraction
  - build-knowledge-graph

This KG plot is based on the first two section of the research paper: "Determinants of longevity: genetic, environmental and medical factors"

https://github.com/thunlp/OpenNRE Comment: Tested the framework on our research paper. We observed poor performance on the scientific paper/excerpt.

https://github.com/princeton-nlp/PURE Comment: Very promising. This project has:

• SciBERT-approx (cross, W=100) (390M): Cross-sentence approximation relation model based on allenai/scibert_scivocab_uncased
• Time permitting, I would like to explore more about the performance. As of now - framework allows only to evaluate the model.

https://arxiv.org/pdf/2210.14698.pdf

Comment: It is a conditional language model that is based on the intra-structure dependencies. Limited open-source to explore how it does on a sample sentence/document.

https://github.com/zjunlp/DeepKE A great collection of NLP related tasks like relation extraction. The downside is limited access to English language pre-trained model. They do provide ways to train your model on custom datasets. It can be handy.

Part of DeepKE: https://github.com/FDUyjx/RSMAN

• https://huggingface.co/datasets/docred
• https://github.com/Babelscape/rebel