Implementation of Our Paper "Controllable Abstractive Sentence Summarization with Guiding Entities" in COLING 2020.

• tensorflow > 2.1.0
• tqdm
• pyrouge
• numpy
• nltk

Our controllable neural model with guiding entities. The original article texts are encoded with a BiLSTM layer. We utilize a pretrained BERT named entity recognition tool to extract entities from input texts. The decoder consists of two LSTMs: LSTM-L and LSTM-R. Our model starts generating the left and right part of a summary with selected entities and can guarantee that entities appear in final output summaries.

Check the README.md file for more details about sources and usage in ./data.

The preprocessed data are composed by three parts:

Source file, each line contains a sentence

Sonia Sotomayor was sworn in Saturday as the Supreme Court 's first Hispanic justice and only third female member in the top U.S. court 's 220-year history.

Target file, each line contains a reference summary

Sotomayor sworn in to top U.S. court.

Object file, each line contains the extracted entities, separated by a delimiter

Sonia Sotomayor <sep> Supreme court <sep> Hispanic <sep> U.S.

Set parameter and path in config.py

Tokenize with tokenizer.py

Train the model with train.py

If you find this repo helpful, please cite the following:

@inproceedings{zheng2020controllable,
  title={Controllable Abstractive Sentence Summarization with Guiding Entities},
  author={Zheng, Changmeng and Cai, Yi and Zhang, Guanjie and Li, Qing},
  booktitle={Proceedings of the 28th International Conference on Computational Linguistics},
  pages={5668--5678},
  year={2020}
}