Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
41 tokens/sec
GPT-4o
59 tokens/sec
Gemini 2.5 Pro Pro
41 tokens/sec
o3 Pro
7 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

SummaReranker: A Multi-Task Mixture-of-Experts Re-ranking Framework for Abstractive Summarization (2203.06569v2)

Published 13 Mar 2022 in cs.CL

Abstract: Sequence-to-sequence neural networks have recently achieved great success in abstractive summarization, especially through fine-tuning large pre-trained LLMs on the downstream dataset. These models are typically decoded with beam search to generate a unique summary. However, the search space is very large, and with the exposure bias, such decoding is not optimal. In this paper, we show that it is possible to directly train a second-stage model performing re-ranking on a set of summary candidates. Our mixture-of-experts SummaReranker learns to select a better candidate and consistently improves the performance of the base model. With a base PEGASUS, we push ROUGE scores by 5.44% on CNN-DailyMail (47.16 ROUGE-1), 1.31% on XSum (48.12 ROUGE-1) and 9.34% on Reddit TIFU (29.83 ROUGE-1), reaching a new state-of-the-art. Our code and checkpoints will be available at https://github.com/ntunlp/SummaReranker.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (3)
  1. Mathieu Ravaut (17 papers)
  2. Shafiq Joty (187 papers)
  3. Nancy F. Chen (97 papers)
Citations (81)
Github Logo Streamline Icon: https://streamlinehq.com