tuananhphan97vn/HeterGraphLongSum

We inherit partly code from https://github.com/dqwang122/HeterSumGraph.

Detail for data format used, dependency, rouge-installation can be found in above link.

The command for training process and evaluation are minimally different with several arguments.

Thanks for their work.

Data Processing

1. Download Pubmed and Arxiv dataset from https://github.com/armancohan/long-summarization

2. Preprocess data

For pubmed dataset:

python preprocess_data.py --input_path dataset/pubmed-dataset --output_path dataset/pubmed --task train
python preprocess_data.py --input_path dataset/pubmed-dataset --output_path dataset/pubmed --task val
python preprocess_data.py --input_path dataset/pubmed-dataset --output_path dataset/pubmed --task test

For arxiv dataset:

python preprocess_data.py --input_path dataset/arxiv-dataset --output_path dataset/arxiv --task train
python preprocess_data.py --input_path dataset/arxiv-dataset --output_path dataset/arxiv --task val
python preprocess_data.py --input_path dataset/arxiv-dataset --output_path dataset/arxiv --task test

After getting the standard json format, you can prepare the dataset for the graph by PrepareDataset.sh in the project directory. The processed files will be put under the cache directory.

Train

For training, you can run commands like this:

python train.py --cuda --gpu 0 --data_dir <data/dir/of/your/json-format/dataset> --cache_dir <cache/directory/of/graph/features> --embedding_path <glove_path> --model [HSG|HDSG] --save_root <model path> --log_root <log path> --lr_descent --grad_clip -m 6 --save_name folder_name --use_doc --n_iter 2 --passage_length 10 --full_data full

## Test

For evaluation, the command may like this:


python evaluation.py --cuda --gpu 0 --data_dir <data/dir/of/your/json-format/dataset> --cache_dir <cache/directory/of/graph/features> --embedding_path <glove_path>  --model [HSG|HDSG] --save_root <model path>  -m 6 --test_model multi --use_pyrouge --passage_length 10 --doc_max_timesteps 150 --n_iter 2 --use_doc  --gpu 0 --batch_size 32


Some options:

- use_doc: whether to use doc_representation for classification
- passage_length: number of sentence to create one passage 
- niter: number iteration for updating value of weight
- *use_pyrouge*: whether to use pyrouge for evaluation. Default is **False** (which means rouge).
  - Please change Line17-18 in ***tools/utils.py*** to your own ROUGE path and temp file path.
- *limit*: whether to limit the output to the length of gold summaries. This option is only set for evaluation on NYT50 (which uses ROUGE-recall instead of ROUGE-f). Default is **False**.
- *blocking*: whether to use Trigram blocking. Default is **False**.
- save_label: only save label and do not calculate ROUGE. Default is **False**.