Obsidian Source: Notes / Relationship Detection
Summary
Pending synthesis from local Obsidian source.
Original source title: Relationship Detection
Extracted Preview
Automating process of detection among entities, and relationship has considerable value. Relationship extraction is the task of extracting semantic relationship from a text. As with NER, RE process also extracts structured information from unstructured text. When text has gone through NER process, RE process can extract graphs(knowledge graphs). It might often be necessary to train your own model with labelled data from our own domain.
!Pasted image 20230801225818.png
Relation Extraction component applied to same sentence.
- The BERT models are there, which can be fine-tuned on specific tasks. https://github.com/sujitpal/ner-re-with-transformers-odsc2022
- Some papers and code for this particular type of problem : https://paperswithcode.com/task/relation-extraction/codeless
- A similar problem, how the data should be there and how to fine tune for your case, and a space with an example : https://huggingface.co/yseop/distilbert-base-financial-relation-extraction
- A notebook implementation, on New York Times Relation Extraction Dataset : https://github.com/sujitpal/ner-re-with-transformers-odsc2022/blob/main/05-nyt-re-bert.ipynb
There might not be any model available for relationship extraction, but the above notebook can be used as a starting point. It basically classify some relation type.
Integration Notes
- Source folder:
/home/yashs/Documents/Docs/Obsidian/Research-Notes - Local source:
/home/yashs/Documents/Docs/Obsidian/Research-Notes/Notes/Relationship Detection.md - Raw copy:
raw/obsidian/research-notes/Notes/Relationship Detection.md