NLP Research

Relation Extraction in Biomedical using BI-LSTM-CRF model + Bert Embedding + Keras + Tensorflow

Slamet Riyanto S.Kom., M.M.S.I.

Dimas Dwi Putra

Architecture

Dataset
View Dataset .txt

Dictionary .json

{"Negative": 0, "Cause_of_disease": 1, "Treatment_of_disease": 2, "Association": 3}

Train Dataset .txt View Directory

2 Lonicera caerulaea berry$solid tumors$The antioxidant response induced by ######################## extracts in animals bearing experimental ############ . 
1 tobacco$cancer$Targeted smoking prevention and cessation interventions are needed for ###### survivors, especially those diagnosed with a ####### -related ###### . 
0 green tea$pneumonia$OBJECTIVE: We examined the association between ######### consumption and death from ######### in humans. 
...

Test Dataset .txt View Directory

1 coffee$addiction$The consumption of ###### is a self-reinforcing behavior, and caffeine dependence and ######### are common. 
0 coffee$coronary disease$Initial trials showing ###### 's association with ################ and myocardial infarction have been difficult to reproduce and have many confounding variables. 
3 coffee$coronary heart disease$BACKGROUND: For many years, an association between ###### consumption and the risk of ###################### has been suspected.
...

Source Code

Utils
View Directory

- args.py
- att.py
- crf.py
- extract_feature.py
- graph.py
- load_data.py
- modeling.py
- optimization.py
- tokenization.py

Notebook

• BILSTM-CRF-RE.ipynb
  
• BILSTM-CRF-ATTENTION-FC-RE

Visualization

Predict
View More Predict Results .xlsx

original: green tea#skin papillomas#Inhibitory effect of green tea on the growth of established skin papillomas in mice.
Predict: Treatment_of_disease

sentences	status	plant	disease	relation
Public awareness about tobacco -related oral cancer is low at present, and new approaches to this problem should include education in the schools on oral cancer, formulation of legislative action to ban the sale of tobacco near schools and colleges, and imposition of societal "barriers" that would make the nonchewing of tobacco socially more acceptable than chewing it.	success	detect	detect	make relation
the mechanism by which magnesium affects digitalis induced arrhythmias was studied in dogs with and without beta receptor	success	detect	detect	make relation
these observations indicate that oral administration of green tea i p a polyphenol fraction or i p epigallocatechin gallate inhibited the growth and caused regression established experimentally induced skin papillomas	success	detect	detect	make relation
...

Eval
View Model Report.xlsx

Entities	precision	recall	f1-score	support	processor	ram	model	batch size	epochs	length	embedding	Uji	excecution time
Negative	0.76	0.78	0.77	118	cpu	high	6	8	71/100	512	128	12	5.15.15
Cause Of Disease	0.68	0.76	0.72	37
Treatment Of Diesease	0.86	0.85	0.85	98
Association	0.50	0.12	0.20	8
micro avg			0.78	261
macro avg	0.70	0.63	0.63	261
weighted avg	0.78	0.78	0.78	261
F-1 Scores			78,16%

Requirements

install from requirements.txt

python==3.6
pandas==0.23.4
keras==2.3.1
termcolor==1.1.0
six==1.16.0
tensorflow==1.13.1
numpy==1.16.2
matplotlib==2.2.4
scikit-learn==0.24.2
h5py==2.10.0
git+https://www.github.com/keras-team/keras-contrib.git

Model Output
View Model Output Directory

create model using keras and tensorflow .hdf5

results/
  uji/.hdf5

Other Content

Websites Prediction

1. Django Websites Prediction For NER dan RE

Named Entity Recognition (NER)

1. NER Dataset Biomedical Plant-Disease Corpus

2. NER CRF Named Entity Recognition

3. NER BiLSTM Named Entity Recognition

4. NER BERT Named Entity Recognition

5. NER BiLSTM CRF Named Entity Recognition

6. NER BERT BiLSTM CRF Named Entity Recognition

Relation Extraction (RE)

1. RE Dataset Biomedical Plant-Disease Corpus

2. RE BERT Relation Extraction Biomedical

3. RE BiLSTM CRF Relation Extraction Biomedical