This paper proposes a model called Edisum that can generate useful edit summaries for Wikipedia edits at scale

This paper proposes a model called Edisum that can generate useful edit summaries for Wikipedia edits at scale. The model is trained on a curated mix of human and synthetic data to overcome the challenges of mixed-quality training data and efficiency requirements imposed by the scale of Wikipedia. The model performs on par with human editors and outperforms open-source language models on this task.