Researchers have introduced LOOMSUM, a novel framework designed to improve the faithfulness of long text-table summarization. This training-free approach focuses on extracting atomic evidence from source documents, explicitly linking quantitative facts from tables with supporting narrative analyses, and planning the discourse structure prior to generation. To evaluate its effectiveness, a new metric called Table-Grounded Faithfulness (TGF) was developed, which assesses numeric grounding, analysis support, and relation consistency at the claim level. Experiments on FINDSum and USTT benchmarks demonstrated that LOOMSUM enhances analytical faithfulness and maintains strong summarization quality, with human evaluations showing positive associations between its components and human judgments. AI
IMPACT Enhances faithfulness in summarizing complex documents with both text and tables, potentially improving information extraction and analysis.
RANK_REASON The cluster contains a research paper detailing a new method and metric for text-table summarization. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
