PreSumm

This code is for EMNLP 2019 paper Text Summarization with Pretrained Encoders

Updates Jan 22 2020: Now you can Summarize Raw Text Input!. Swith to the dev branch, and use -mode test_text and use -text_src $RAW_SRC.TXT to input your text file. Please still use master branch for normal training and evaluation, dev branch should be only used for test_text mode.

• abstractive use -task abs, extractive use -task ext
• use -test_from $PT_FILE$ to use your model checkpoint file.
• Format of the source text file:
  • For abstractive summarization, each line is a document.
  • If you want to do extractive summarization, please insert [CLS] [SEP] as your sentence boundaries.
• There are example input files in the raw_data directory
• If you also have reference summaries aligned with your source input, please use -text_tgt $RAW_TGT.TXT to keep the order for evaluation.

Results on CNN/DailyMail (20/8/2019):

Models	ROUGE-1	ROUGE-2	ROUGE-L
Extractive
TransformerExt	40.90	18.02	37.17
BertSumExt	43.23	20.24	39.63
BertSumExt (large)	43.85	20.34	39.90
Abstractive
TransformerAbs	40.21	17.76	37.09
BertSumAbs	41.72	19.39	38.76
BertSumExtAbs	42.13	19.60	39.18

Python version: This code is in Python3.6

Package Requirements: torch==1.1.0 pytorch_transformers tensorboardX multiprocess pyrouge

Updates: For encoding a text longer than 512 tokens, for example 800. Set max_pos to 800 during both preprocessing and training.

Some codes are borrowed from ONMT(https://github.com/OpenNMT/OpenNMT-py)

Trained Models

CNN/DM BertExt

CNN/DM BertExtAbs

CNN/DM TransformerAbs

XSum BertExtAbs

System Outputs

CNN/DM and XSum

Data Preparation For XSum

Pre-processed data

Data Preparation For CNN/Dailymail

Option 1: download the processed data

Pre-processed data

unzip the zipfile and put all .pt files into bert_data

Option 2: process the data yourself

Step 1 Download Stories

Download and unzip the stories directories from here for both CNN and Daily Mail. Put all .story files in one directory (e.g. ../raw_stories)

Step 2. Download Stanford CoreNLP

We will need Stanford CoreNLP to tokenize the data. Download it here and unzip it. Then add the following command to your bash_profile:

export CLASSPATH=/path/to/stanford-corenlp-full-2017-06-09/stanford-corenlp-3.8.0.jar

replacing /path/to/ with the path to where you saved the stanford-corenlp-full-2017-06-09 directory.

Step 3. Sentence Splitting and Tokenization

python preprocess.py -mode tokenize -raw_path RAW_PATH -save_path TOKENIZED_PATH

• RAW_PATH is the directory containing story files (../raw_stories), TOKENIZED_PATH is the target directory to save the generated json files (../merged_stories_tokenized)

Step 4. Format to Simpler Json Files

python preprocess.py -mode format_to_lines -raw_path RAW_PATH -save_path JSON_PATH -n_cpus 1 -use_bert_basic_tokenizer false -map_path MAP_PATH

• RAW_PATH is the directory containing tokenized files (../merged_stories_tokenized), JSON_PATH is the target directory to save the generated json files (../json_data/cnndm), MAP_PATH is the directory containing the urls files (../urls)

Step 5. Format to PyTorch Files

python preprocess.py -mode format_to_bert -raw_path JSON_PATH -save_path BERT_DATA_PATH  -lower -n_cpus 1 -log_file ../logs/preprocess.log

• JSON_PATH is the directory containing json files (../json_data), BERT_DATA_PATH is the target directory to save the generated binary files (../bert_data)

Option 3: Process your own data(jsonl)

Step 1. Download and move the jsonl file to datasets/YOUR_DATASET_NAME/raw

• CORNELL NEWSROOM

Step 2. Setting for using Stanford CoreNLP

• Stanford CoreNLP
• Stanford CoreNLP Python Interface

# Setting Stanford CoreNLP
cd ~
wget http://nlp.stanford.edu/software/stanford-corenlp-4.2.0.zip
unzip stanford-corenlp-4.2.0.zip && cd stanford-corenlp-4.2.0
pip install-U stanford-corenlp

export CORENLP_HOME="~/stanford-corenlp-4.2.0"

# Install required libraries for preprocessing
pip install -r requirements_prepro.txt

Step 3. Processing(jsonl_to_df)

• This process run the local server for CoreNLP process(tokenizer) half the number of n_cpus.
  So set the -n_cpus value below than (the number of your cpu core) * 2/3.

# Step by step
python preprocess.py -mode jsonls_to_dfs -dataset YOUR_DATASET_NAME -n_cpus CPU_NUM
python preprocess.py -mode dfs_to_jsons -dataset YOUR_DATASET_NAME -n_cpus CPU_NUM

Or

# Do it all at once
python preprocess.py -mode jsonl_to_bert -dataset YOUR_DATASET_NAME -n_cpus CPU_NUM

Model Training

First run: For the first time, you should use single-GPU, so the code can download the BERT model. Use -visible_gpus -1, after downloading, you could kill the process and rerun the code with multi-GPUs.

Extractive Setting

python train.py -task ext -mode train -bert_data_path BERT_DATA_PATH -ext_dropout 0.1 -model_path MODEL_PATH -lr 2e-3 -visible_gpus 0,1,2 -report_every 50 -save_checkpoint_steps 1000 -batch_size 3000 -train_steps 50000 -accum_count 2 -log_file ../logs/ext_bert_cnndm -use_interval true -warmup_steps 10000 -max_pos 512

Abstractive Setting

TransformerAbs (baseline)

python train.py -mode train -accum_count 5 -batch_size 300 -bert_data_path BERT_DATA_PATH -dec_dropout 0.1 -log_file ../../logs/cnndm_baseline -lr 0.05 -model_path MODEL_PATH -save_checkpoint_steps 2000 -seed 777 -sep_optim false -train_steps 200000 -use_bert_emb true -use_interval true -warmup_steps 8000  -visible_gpus 0,1,2,3 -max_pos 512 -report_every 50 -enc_hidden_size 512  -enc_layers 6 -enc_ff_size 2048 -enc_dropout 0.1 -dec_layers 6 -dec_hidden_size 512 -dec_ff_size 2048 -encoder baseline -task abs

BertAbs

python train.py  -task abs -mode train -bert_data_path BERT_DATA_PATH -dec_dropout 0.2  -model_path MODEL_PATH -sep_optim true -lr_bert 0.002 -lr_dec 0.2 -save_checkpoint_steps 2000 -batch_size 140 -train_steps 200000 -report_every 50 -accum_count 5 -use_bert_emb true -use_interval true -warmup_steps_bert 20000 -warmup_steps_dec 10000 -max_pos 512 -visible_gpus 0,1,2,3  -log_file ../logs/abs_bert_cnndm

BertExtAbs

python train.py  -task abs -mode train -bert_data_path BERT_DATA_PATH -dec_dropout 0.2  -model_path MODEL_PATH -sep_optim true -lr_bert 0.002 -lr_dec 0.2 -save_checkpoint_steps 2000 -batch_size 140 -train_steps 200000 -report_every 50 -accum_count 5 -use_bert_emb true -use_interval true -warmup_steps_bert 20000 -warmup_steps_dec 10000 -max_pos 512 -visible_gpus 0,1,2,3 -log_file ../logs/abs_bert_cnndm  -load_from_extractive EXT_CKPT   

• EXT_CKPT is the saved .pt checkpoint of the extractive model.

Model Evaluation

CNN/DM

 python train.py -task abs -mode validate -batch_size 3000 -test_batch_size 500 -bert_data_path BERT_DATA_PATH -log_file ../logs/val_abs_bert_cnndm -model_path MODEL_PATH -sep_optim true -use_interval true -visible_gpus 1 -max_pos 512 -max_length 200 -alpha 0.95 -min_length 50 -result_path ../logs/abs_bert_cnndm 

XSum

 python train.py -task abs -mode validate -batch_size 3000 -test_batch_size 500 -bert_data_path BERT_DATA_PATH -log_file ../logs/val_abs_bert_cnndm -model_path MODEL_PATH -sep_optim true -use_interval true -visible_gpus 1 -max_pos 512 -min_length 20 -max_length 100 -alpha 0.9 -result_path ../logs/abs_bert_cnndm 

• -mode can be {validate, test}, where validate will inspect the model directory and evaluate the model for each newly saved checkpoint, test need to be used with -test_from, indicating the checkpoint you want to use
• MODEL_PATH is the directory of saved checkpoints
• use -mode valiadte with -test_all, the system will load all saved checkpoints and select the top ones to generate summaries (this will take a while)