- Collecting datasets for ABS is challenging due to the lack of available summaries.
- Unlike generic summarization, ABS requires writing high-quality aspect-based summaries from scratch or crowdsourcing, which is expensive and time-consuming.
- Existing datasets like AnyAspect and OASUM are synthetically compiled and only address single document inputs.
- These methods yield lower-quality input documents, aspects, and expected output summaries.
- A new direction in ABS allows open aspects that concisely target subtopics but can be unique for individual input text.
- More datasets and benchmarks are needed to enable research on these tasks.
Overall, the article emphasizes the difficulties of collecting high-quality aspect-based summaries for research purposes. It highlights the need for more diverse and comprehensive datasets that can help improve the performance of ABS models. By providing a detailed overview of the challenges involved in creating such datasets, the authors encourage further research in this area to advance the field of ABS.
