Factorizing Content and Budget Decisions in Abstractive Summarization of Long Documents

TL;DR

FactorSum introduces a novel approach to long document summarization by disentangling content selection from budget constraints, leading to improved ROUGE scores and better domain adaptation.

Contribution

This work presents FactorSum, a new method that factorizes summarization into content and budget steps, enhancing performance and flexibility over existing models.

Findings

Achieves higher ROUGE scores on PubMed, arXiv, and GovReport benchmarks.

Effective domain adaptation with only PubMed training data.

Flexible budget and content guidance improve summarization quality.

Abstract

We argue that disentangling content selection from the budget used to cover salient content improves the performance and applicability of abstractive summarizers. Our method, FactorSum, does this disentanglement by factorizing summarization into two steps through an energy function: (1) generation of abstractive summary views; (2) combination of these views into a final summary, following a budget and content guidance. This guidance may come from different sources, including from an advisor model such as BART or BigBird, or in oracle mode -- from the reference. This factorization achieves significantly higher ROUGE scores on multiple benchmarks for long document summarization, namely PubMed, arXiv, and GovReport. Most notably, our model is effective for domain adaptation. When trained only on PubMed samples, it achieves a 46.29 ROUGE-1 score on arXiv, which indicates a strong performance due to more flexible budget adaptation and content selection less dependent on domain-specific textual structure.