Skip to content

Code for "Multi-document Summarization with Maximal Marginal Relevance-guided Reinforcement Learning", EMNLP 2020

Notifications You must be signed in to change notification settings

morningmoni/RL-MMR

Repository files navigation

[update 2021/06] The system summaries are uploaded as output.zip.

Code

train_full_rl.py: Main file. Call training.py -> rl.py -> model/rl.py.

training.py: training loop

rl.py: training epoch (rewards, official evaluation)

model/rl.py: core functions, where MMR is injected

ConfManager.py: Parameters in addition to argparse

data_info.py: store all data paths, data follows the format of fast_abs_rl

ScoreAgent.py: MMR

Acknowledgments

Part of the code is adapted from fast_abs_rl.

About

Code for "Multi-document Summarization with Maximal Marginal Relevance-guided Reinforcement Learning", EMNLP 2020

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages