Text Summarization Papers

by Pengfei Liu, Yiran Chen, Jinlan Fu, Hiroaki Hayashi, Danqing Wang and other contributors.

An exhaustive paper list for Text Summarization, covering papers from eight top conferences (ACL / EMNLP / NAACL / ICML / ICLR / AAAI / IJCAI / NeurIPS) in the last eight years (2013-2020).

What can I get here?

1. Paper Retrieval System [Click Me !!!] ?

  • Find the top-cited summarization papers! [The latest update on: 02.25/2020]
  • Track the latest summarization papers!
  • Find the milestone summarization papers for beginners.
  • Search papers by research concepts or your interested keywords.

2. What are the recent Research Concepts and which are HOT?

We first define the typology of essential concepts for the summarization task. We then plot the number of papers for each concept below.

Text-Summarization-Papers

Trends in 2019:

Concepts in red suggest HOT topics, and we can observe:

  • Task: Scientific paper-based summarization has gain growing interests.
  • Data: More new datasets are constructed.
  • Architecture: Pretrained models and graph neural networks prevail.
  • Evaluation: Evaluation of the generated summary's factuality attracts recent attention.
Hot topic: when the proportion of papers on a concept since 2019 is greater than a certain threshold (0.4), we define this concept as a hot topic.

3. Recommended Papers

Papers with Hot Topics

  • pre-X: summarizer with unsupervised pretrained models.
  • task-sci: scientifc paper-based summarization.
  • eval-factuality: factuality evaluation on generated summaries.
  • arch-gnn: graph neural network-based summarizers.
  • data-new: more new datasets are constructed.

Milestone Papers

4. Mainstream Dataset List ?

What can I do here?

  • If you have a new "research concept" -- Tell us

    • Update the file summ_concept.md and send us a Pull request.
    • Or you could open an Issue.

  • If you have a new "paper" or want to modify our inaccurate annotations of concepts:

    • Update your paper into the file summ_paper.crowdsource and send us a Pull request.
    • Or you could open an Issue.

  • If you have a new "dataset" or want to modify our inaccurate annotations:

    • Add your dataset (If possible, with a brief description) into summ_data.crowdsource and send us a Pull request.
    • Or you could open an Issue.

Related Work

Future Work

Hopefully, you will see our version-2.0 covering papers from 1980 to 2020.

GitHub