Source code for "ECNU-SenseMaker at SemEval-2020 Task 4: Leveraging Heterogeneous Knowledge Resources for Commonsense Validation and Explanation" (SemEval 2020).
-
The overview of ECNU-SenseMaker.
-
The KEmb module.
| User | Team Name | Accuracy | Rank |
|---|---|---|---|
| im0qianqian | ECNU_ICA | 95.0 | 1 |
| jession | hit_itnlp | 94.8 | 2 |
| Solomon | Solomon | 94.0 | 3 |
| w-h-m | NEUKG | 93.8 | 3 |
| xlx | NUT | 93.7 | 5 |
| User | Team Name | Accuracy | Rank |
|---|---|---|---|
| jession | hit_itnlp | 97.0 | 1 |
| im0qianqian | ECNU_ICA | 96.7 | 2 |
| xlx | NUT | 96.4 | 3 |
| nlpx | 96.4 | 3 | |
| Solomon | Solomon | 96.0 | 5 |
# create a conda envs
conda create -n semeval2020 python=3.6 ipython
conda activate semeval2020
# pytorch
conda install pytorch=1.6 cudatoolkit=10.2 -c pytorch
# transformers
pip install transformers==2.5.1
# pyG
CUDA=cu102
TORCH=1.6.0
pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-${TORCH}+${CUDA}.html
pip install torch-sparse -f https://pytorch-geometric.com/whl/torch-${TORCH}+${CUDA}.html
pip install torch-cluster -f https://pytorch-geometric.com/whl/torch-${TORCH}+${CUDA}.html
pip install torch-spline-conv -f https://pytorch-geometric.com/whl/torch-${TORCH}+${CUDA}.html
pip install torch-geometric
# sklearn
conda install scikit-learn=0.21.3
# bidict
pip install bidict==0.18.3
# nltk
conda install nltk
# tensorboard
conda install TensorBoard==1.15
# others
pip install future
pip install wordfreq- Download the
conceptnet5.zipfrom here, and unzip it to theconceptnet5/directory. - Download the
pre_weights.zipfrom here, and unzip it to thepre_weights/directory.
The directory tree of ECNU-SenseMaker:
ECNU-SenseMaker
โโโ commonsenseQA
โย ย โโโ dev_rand_split_EASY.jsonl
โย ย โโโ dev_rand_split.jsonl
โย ย โโโ test_rand_split_no_answers.jsonl
โย ย โโโ train_rand_split_EASY.jsonl
โย ย โโโ train_rand_split.jsonl
โโโ conceptnet5
โย ย โโโ numberbatch-en.txt
โย ย โโโ res_all.pickle
โโโ config.py
โโโ functions.py
โโโ loss.py
โโโ model_modify.py
โโโ models.py
โโโ optimizer.py
โโโ pre_weights
โย ย โโโ albert-xxlarge_model.bin
โย ย โโโ bert-base-uncased_model.bin
โย ย โโโ roberta-base_model.bin
โย ย โโโ roberta-large_model.bin
โโโ README.md
โโโ run_ensemble_model.py
โโโ run_single_model.py
โโโ SemEval2020-Task4-Commonsense-Validation-and-Explanation-master
โย ย โโโ Training Data
โย ย โย ย โโโ subtaskA_answers_all.csv
โย ย โย ย โโโ subtaskA_data_all.csv
โย ย โย ย โโโ subtaskB_answers_all.csv
โย ย โย ย โโโ subtaskB_data_all.csv
โย ย โย ย โโโ subtaskC_answers_all.csv
โย ย โย ย โโโ subtaskC_data_all.csv
โย ย ...
|
โโโ utils
โโโ attentionUtils.py
โโโ commonsenseQAutils.py
โโโ ensembleUtils.py
โโโ getGraphUtils.py
โโโ gpu_mem_track.py
โโโ GraphUtils.py
โโโ __init__.py
โโโ MyDataset.py
โโโ semevalUtils.py
โโโ testUtils.py
โโโ text_to_uri.py
Run example on Subtask B with LM and KEGAT:
CUDA_VISIBLE_DEVICES='0' python -u -m run_single_model \
--batch-size 8 \
--test-batch-size 8 \
--epochs 4 \
--fine-tune-epochs 8 \
--lr 0.001 \
--fine-tune-lr 0.000005 \
--adam-epsilon 0.000001 \
--max-seq-length 128 \
--subtask-id B \
--with-lm \
--with-kegat
Options of run_single_model.py:
usage: run_single_model.py [-h] [--batch-size N] [--test-batch-size N]
[--epochs N] [--fine-tune-epochs N] [--lr LR]
[--fine-tune-lr LR] [--adam-epsilon M]
[--max-seq-length N] [--subtask-id {A,B}]
[--with-lm] [--with-kegat] [--with-kemb]
[--no-cuda] [--seed S]
ECNU-SenseMaker single model
optional arguments:
-h, --help show this help message and exit
--batch-size N input batch size for training (default: 8)
--test-batch-size N input batch size for testing (default: 8)
--epochs N number of epochs to train (default: 4)
--fine-tune-epochs N number of fine-tune epochs to train (default: 8)
--lr LR learning rate (default: 0.001)
--fine-tune-lr LR fine-tune learning rate (default: 5e-06)
--adam-epsilon M Adam epsilon (default: 1e-06)
--max-seq-length N max length of sentences (default: 128)
--subtask-id {A,B} subtask A or B (default: B)
--with-lm Add Internal Sharing Mechanism (LM)
--with-kegat Add Knowledge-enhanced Graph Attention Network (KEGAT)
--with-kemb Add Knowledge-enhanced Embedding (KEmb)
--no-cuda disables CUDA training
--seed S random seed (default: 1)
Please cite this repository using the following reference:
@inproceedings{zhao-etal-2020-ecnu,
title = "{ECNU}-{S}ense{M}aker at {S}em{E}val-2020 Task 4: Leveraging Heterogeneous Knowledge Resources for Commonsense Validation and Explanation",
author = "Zhao, Qian and
Tao, Siyu and
Zhou, Jie and
Wang, Linlin and
Lin, Xin and
He, Liang",
booktitle = "Proceedings of the Fourteenth Workshop on Semantic Evaluation",
month = dec,
year = "2020",
address = "Barcelona (online)",
publisher = "International Committee for Computational Linguistics",
url = "https://www.aclweb.org/anthology/2020.semeval-1.48",
pages = "401--410",
abstract = "This paper describes our system for SemEval-2020 Task 4: Commonsense Validation and Explanation (Wang et al., 2020). We propose a novel Knowledge-enhanced Graph Attention Network (KEGAT) architecture for this task, leveraging heterogeneous knowledge from both the structured knowledge base (i.e. ConceptNet) and unstructured text to better improve the ability of a machine in commonsense understanding. This model has a powerful commonsense inference capability via utilizing suitable commonsense incorporation methods and upgraded data augmentation techniques. Besides, an internal sharing mechanism is cooperated to prohibit our model from insufficient and excessive reasoning for commonsense. As a result, this model performs quite well in both validation and explanation. For instance, it achieves state-of-the-art accuracy in the subtask called Commonsense Explanation (Multi-Choice). We officially name the system as ECNU-SenseMaker. Code is publicly available at https://github.com/ECNU-ICA/ECNU-SenseMaker.",
}
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent