Skip to content

Relation Extraction

Introduction

  • Relation extraction (RE) is the process of identifying the relationships between entities in a text. Entities could be of multiple types such as person, location, organization, etc and they can be identified using Named Enitity Recognition (NER).
  • Let's understand RE with an example: "Ram is the son of Shyam and Shyam is the son of Radhe". Here the entities are identified as: "Ram", "Shyam" and "Radhe". The relations could be (Ram, son of, Shyam), (Shyam, son of, Radhe) and (Ram, grandson of, Radhe).

Code

Using OpenNRE

  • OpenNRE is an open source tool for relationship extraction. OpenNRE makes it easy to extract relationships from text. It is as simple as writing a few lines of code.
  • One example from their github repository is as follows, 
1
2
3
4
5
6
7
# import opennre
import opennre
# load the model
model = opennre.get_model('wiki80_cnn_softmax')
# infer for a text
model.infer({'text': 'He was the son of Máel Dúin mac Máele Fithrich, and grandson of the high king Áed Uaridnach (died 612).', 'h': {'pos': (18, 46)}, 't': {'pos': (78, 91)}})
# Output: ('father', 0.5108704566955566)
  • At the time of writing they had following models available:
model_name description
wiki80_cnn_softmax trained on wiki80 dataset with a CNN encoder.
wiki80_bert_softmax trained on wiki80 dataset with a BERT encoder.
wiki80_bertentity_softmax trained on wiki80 dataset with a BERT encoder (using entity representation concatenation).
tacred_bert_softmax trained on TACRED dataset with a BERT encoder.
tacred_bertentity_softmax trained on TACRED dataset with a BERT encoder (using entity representation concatenation).