Retrieved-Span Training for Efficient Query-Focused Meeting Summarization on QMSum

基于检索片段训练的 QMSum 高效查询聚焦会议摘要研究

Abstract: QMSum provides no scorer, making query-focused meeting summarization results difficult to compare. We rescore or generate 15 systems under one implementation. Through a common inference port, a released 406M Fusion-in-Decoder specialist loses 6.30 ROUGE-1 when moved from capped long input to 2,000-word retrieved spans. Fine-tuning it on this span regime recovers the loss.

摘要： QMSum 数据集未提供评分器，导致查询聚焦会议摘要的结果难以进行比较。我们在统一的实现框架下对 15 个系统进行了重新评分或生成。通过通用的推理接口，一个已发布的 406M 参数 Fusion-in-Decoder 专家模型在从截断的长文本输入切换到 2,000 词的检索片段时，ROUGE-1 分数下降了 6.30。通过在该片段机制下进行微调，模型恢复了这一性能损失。

On test it scores 36.33 ROUGE-1 versus 35.41 for our 1.2B system; the meeting-cluster 95% interval for the difference is [-0.27, +2.22], so QMSum does not statistically separate them. The smaller system uses about one-third as many total parameters and less than half the peak inference memory.

在测试集上，该模型得分为 36.33 ROUGE-1，而我们的 1.2B 参数系统得分为 35.41；会议聚类的 95% 置信区间差异为 [-0.27, +2.22]，因此 QMSum 在统计学上无法区分两者。较小的系统使用的总参数量约为后者的三分之一，推理峰值内存占用不到一半。

Within the fixed 1.2B base, span-regime fine-tuning adds 5.29 [+4.02, +6.56], while replacing the first 4,500 transcript words with 2,000 retrieved words adds 1.55 on test and 0.29 on validation. Separately, under one concise prompt and reference-overlap scorer, a released 406M specialist exceeds five proprietary hosted models by at least 6.2 ROUGE-1, but output length and absent human or factuality evaluation limit this ordering. Conclusions are limited to QMSum and automatic metrics.

在固定的 1.2B 基座模型中，片段机制微调提升了 5.29 [+4.02, +6.56] 的分数；而将转录文本的前 4,500 个词替换为 2,000 个检索词，在测试集上提升了 1.55 分，在验证集上提升了 0.29 分。此外，在统一的简洁提示词和参考重叠评分器下，一个已发布的 406M 专家模型在 ROUGE-1 指标上至少超过了五个专有托管模型 6.2 分，但输出长度以及缺乏人工或事实性评估限制了这一排名的参考价值。结论仅限于 QMSum 数据集和自动评估指标。