Hypernym and Hyponym Detection Based on Auxiliary Sentences and the BERT Model

Introduction

We utilize English BERT pre-trained model and auxiliary sentences for Hypernym and Hyponym Detection. Paper

Run

Training Task1

• First, go to the path

cd Hypernym-and-Hyponym-Detection-Based-on-Auxiliary-Sentences-and-the-BERT-Model/code/train

• Use

$python task1_BERT.py

or

$python task1_Q.py

or

$python task1_PosNeg.py

to train task1 model with default parameters.

• Use

$python task1_BERT.py -dataset        <path to training set, default to WordNet training set> 
                      -label          <training set with label or not> 
                      -b              <batch size, default 64> 
                      -epoch          <number of epochs, default 19> 
                      -lr             <learning rate, default 1e-5> 
                      -model          <name of model default save to'output/task1_BERT_epoch.pth'> 

to train with your own parameters.

Training Task2

• First, go to the path

cd code
cd train

• Use

$python task2_BERT.py

or

$python task2_Q.py

or

$python task2_PosNeg.py

to train task2 model with default parameters.

• Use

$python task2_BERT.py -dataset        <path to training set, default to WordNet training set> 
                      -label          <training set with label or not> 
                      -b              <batch size, default 64> 
                      -epoch          <number of epochs, default 19> 
                      -lr             <learning rate, default 1e-5> 
                      -model          <name of model default save to'output/task1_BERT_epoch.pth'> 

to train with your own parameters.

• Use create_negative_dataset.py to generate task1 negative dataset for PosNeg model.

Evaluation

• First, go to the path

cd code
cd evaluate

Task1 Evaluation

• Use

$python eval_task1_BERT.py -model_path MODEL_PATH    <path to model>

$python eval_task1_Q.py -model_path MODEL_PATH    <path to model>

$python eval_task1_PosNeg.py -model_path MODEL_PATH    <path to model>

to evaluate task1 model.

Eask2 and Eask1&2 Evaluation

• Use

$python eval_task1_2_BERT.py -model1_path <MODEL1_PATH>
                             -model2_path <MODEL2_PATH>

$python eval_task1_2_Q.py -model1_path <MODEL1_PATH>
                          -model2_path <MODEL2_PATH>

$python eval_task1_2_PosNeg.py -model1_path <MODEL1_PATH>
                               -model2_path <MODEL2_PATH>

to evaluate task2 and task1&2.

Datasets

• see more in training dataset README