DETQUS: Decomposition-Enhanced Transformers for QUery-focused Summarization
Published in Proceedings of the 2025 Conference of the NAACL: Human Language Technologies (Volume 1: Long Papers), 2025
Recommended citation: Khan, Y., Wu, X., Youm, S., Ho, J., Shaikh, A., Garciga, J., Sharma, R., & Dorr, B. (2025). DETQUS: Decomposition-Enhanced Transformers for QUery-focused Summarization. In Proceedings of the NAACL 2025: Human Language Technologies, Vol. 1. https://aclanthology.org/2025.naacl-long.138/
Query-focused tabular summarization is a challenging table-to-text task where summaries must answer user queries by extracting relevant content from large tables. DETQUS proposes a decomposition-based transformer system that reduces table complexity while preserving key information. By pruning irrelevant columns and applying a fine-tuned encoder-decoder model, DETQUS improves summarization efficiency and achieves a ROUGE-L score of 0.4437, outperforming the state-of-the-art REFACToR baseline. This paper presents DETQUS as a scalable and interpretable solution for real-world query-focused summarization.