search

HenryJunW / ESSumm

Apache License 2.0
6 stars 0 forks source link

the labels for extractive summarization #1

Closed qipotianMFXT closed 1 year ago

qipotianMFXT commented 1 year ago

Excuse me, your paper is about end-to-end extractive summarization, but in the "data" folder, only the labels for abstractive summarization can be found. Are you sure that you evaluated the performance of extractive summarization using the labels for abstractive summarization, or is it possible that the labels for extractive summarization have not been updated in the code? In addition, there is no feature extraction code in ESSumm_wav2vec.py, so it seems that this code has not been fully updated. Could you please provide an update?

HenryJunW commented 1 year ago

Hello, thanks for your interest in ESSumm.

  1. Yes, you are right. ESSumm is an extractive summarization method. However, to have a fair comparison with the existing approaches, we compare the generated extractive summary to the ground truth abstractive summary.

  2. The code for the Wav2Vec-based feature extraction is in https://github.com/HenryJunW/ESSumm/blob/a4583d641bbb104a6b7d0d7c8c13c7ecde7116f5/data/ESSumm_utterance_community_detection.py#L161. Please refer to https://github.com/HenryJunW/ESSumm/blob/main/GETTING_STARTED.md, and it's part of 'Segments Generation & Key-segments Extraction'.