Graph Fusion Network for Text Classification（GFN）

The source code of our paper(Graph Fusion Network for Text Classification) presents in KBS 2022. https://doi.org/10.1016/j.knosys.2021.107659

dataset process

we can use process all datasets:

python pre_processing.py python pmi.py

train GFN model

we used four datasets to train GFN model: mr r8 r52 oh

we can use the command to train GFN model, all parameters can be set。

Example for train MR dataset: mr edges: 4

python nni_train_ada_regu.py --dataset mr --edges 4 --global_add_local 1 --global_edge 1 --adopt_nni 0 --adopt_nni_paras 0 --max_epoch 8 --adapte_edge 1

Example for train R8 dataset: r8 edges: 4

python nni_train_ada_regu.py --dataset r8 --edges 4 --global_add_local 1 --global_edge 1 --adopt_nni 0 --adopt_nni_paras 0 --max_epoch 8 --adapte_edge 1

Example for train R52 dataset: r52 edges: 4

python nni_train_ada_regu.py --dataset r52 --edges 4 --global_add_local 1 --global_edge 1 --adopt_nni 0 --adopt_nni_paras 0 --max_epoch 8 --adapte_edge 1

Example for train OH dataset: oh edges: 4

python nni_train_ada_regu.py --dataset oh --edges 4 --global_add_local 1 --global_edge 1 --adopt_nni 0 --adopt_nni_paras 0 --max_epoch 8 --adapte_edge 1

inference GFN model

Example for inference MR dataset:

python learning_and_infer.py --dataset mr --ensemble_model 2 --out_channels 3 --pooler 2 --top_k 1

Example for inference R8 dataset:

python learning_and_infer.py --dataset r8 --ensemble_model 2 --out_channels 3 --pooler 2 --top_k 1

Example for inference R52 dataset:

python learning_and_infer.py --dataset r52 --ensemble_model 2 --out_channels 3 --pooler 2 --top_k 1

Example for inference OH dataset:

python learning_and_infer.py --dataset oh --ensemble_model 2 --out_channels 3 --pooler 2 --top_k 1

file description

config.py : the model configure parameters
pre_processing.py : the data preprocessing
data_helper.py : the data process
pmi.py : build PMI data in passages and generate distance matrix
nni_train_ada_regu.py : train GFN model with adapted edges and regularizer
learning_and_infer.py : inference GFN model in datasets
model_ada.py : the GFN model with adapted
transformer_20ng.py : test 20ng dataset in BERT model
transformer_mr.py : test mr dataset in BERT model
transformer_oh.py : test oh dataset in BERT model
transformer_test.py : test test dataset in BERT model

other note

Note: You need to put a pre-trained Glove model file named glove.6B.300d.txt in the root directory. You get the file from https://nlp.stanford.edu/projects/glove/.

Reference

If you find this project useful, please use the following format to cite the paper:

@article{dai2022graph,
  title={Graph Fusion Network for Text Classification},
  author={Dai, Yong and Shou, Linjun and Gong, Ming and Xia, Xiaolin and Kang, Zhao and Xu, Zenglin and Jiang, Daxin},
  journal={Knowledge-Based Systems},
  volume={236},
  pages={107659},
  year={2022},
  publisher={Elsevier}
}

Contact information

For help or issues, please submit a GitHub issue.

xhjcxxl / GFN