Let Topic Flow: A Unified Topic-guided Segment-wise Dialogue Summarization Framework

This is the repo for TGDS.

We release the initial version of our model's code for immediate access and reproduction. We will subsequently organize and update our code further to facilitate the researchers reading and making incremental experiments.

You may notice that the versions of some packages differ in the requirements.txt files under different folders. This is because we have been consistently conducting experiments with LLMs and some baselines on different GPUs. We plan to standardize them later.

Acknowledgements

The work of this project has been inspired and assisted by the following open-source projects and technologies. We would like to express our gratitude to the developers and contributors of these projects, including but not limited to:

• BERTAbs: https://github.com/nlpyang/PreSumm
• RODS: https://github.com/xiaolinAndy/RODS
• Firefly: https://github.com/yangjianxin1/Firefly

Citation

If you find our work useful, please consider citing it as:

@ARTICLE{10461119,
  author={Han, Qinyu and Yang, Zhihao and Lin, Hongfei and Qin, Tian},
  journal={IEEE/ACM Transactions on Audio, Speech, and Language Processing}, 
  title={Let Topic Flow: A Unified Topic-Guided Segment-Wise Dialogue Summarization Framework}, 
  year={2024},
  volume={32},
  number={},
  pages={2021-2032},
  keywords={Pediatrics;Oral communication;Medical services;Customer services;Task analysis;Indexes;Speech processing;Abstractive dialogue summarization;unified summarization framework;topic-guided segment-wise attention},
  doi={10.1109/TASLP.2024.3374112}}