Disjoint-DABS: A Benchmark for Dynamic Aspect-Based Summarization in Disordered Texts

This is the dataset for the paper \"Disjoint-DABS: A Benchmark for Dynamic Aspect-Based Summarization in Disorganized Texts\". It includes two sub-datasets converted from CNN/DailyMail (D-CnnDM.zip) and WikiHow (D-WikiHow.zip). We include the data with training, validation, and test split. The file for training the summarization model is at (WikiHowSep.zip and CnnDM.zip) We also include the small-scale data for D-WikiHow used for prompting experiments (D-WikiHow-sample). The generated summaries for all baselines for further research, especially for human evaluation is included (result.zip).