Evidence-Based Query Generation for Query-Focused Summarization
A new arXiv paper (2605.05392) proposes an evidence-based model to automatically generate query keywords from query-free summarization datasets for Query-Focused Summarization (QFS). The authors address two research questions: whether evidence-based query keywords can be generated from query-free datasets, and whether such queries support QFS. They evaluate their model intrinsically by comparing original and system-generated queries on two QFS datasets, and extrinsically by performing summarization tasks with various pre-trained models and a state-of-the-art QFS model. Experimental results show that summaries using evidence-based queries achieve competitive ROUGE scores.
Key facts
- Paper proposes evidence-based model for query generation from query-free datasets.
- Addresses two research questions on query generation and QFS support.
- Intrinsic evaluation compares original vs. system-generated queries on two QFS datasets.
- Extrinsic evaluation uses pre-trained models and a SOTA QFS model.
- Summaries with evidence-based queries achieve competitive ROUGE scores.
- Paper is on arXiv with ID 2605.05392.
- Published as arXiv:2605.05392v1.
- Focuses on Query-Focused Summarization (QFS).
Entities
Institutions
- arXiv