This is the official code repository for Exploring Neural Models for Query-Focused Summarization by Jesse Vig*, Alexander R. Fabbri*, Wojciech Kryściński*, Chien-Sheng Wu, and Wenhao Liu (*equal contribution).
We present code and instructions for reproducing the paper experiments and running the models against your own datasets.
Query-focused summarization (QFS) aims to produce summaries that answer particular questions of interest, enabling greater user control and personalization. In our paper we conduct a systematic exploration of neural approaches to QFS, considering two general classes of methods: two-stage extractive-abstractive solutions and end-to-end models. Within those categories, we investigate existing methods and present two model extensions that achieve state-of-the-art performance on the QMSum dataset by a margin of up to 3.38 ROUGE-1, 3.72 ROUGE-2, and 3.28 ROUGE-L.
Two-step approaches consist of an extractor model, which extracts parts of the source document relevant to the input query, and an abstractor model, which synthesizes the extracted segments into a final summary.
See extractors directory for instructions and code for training and evaluating two-stage models.
The Segment Encoder is an end-to-end model that uses sparse local attention to achieve SOTA ROUGE scores on the QMSum dataset.
To replicate the QMSum experiments, or train and evaluate Segment Encoder on your own dataset, see the multiencoder directory.
When referencing this repository, please cite this paper:
@misc{vig-etal-2021-exploring,
title={Exploring Neural Models for Query-Focused Summarization},
author={Jesse Vig and Alexander R. Fabbri and Wojciech Kry{\'s}ci{\'n}ski and Chien-Sheng Wu and Wenhao Liu},
year={2021},
eprint={2112.07637},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2112.07637}
}This repository is released under the BSD-3 License.