A new evaluation framework with XPath-based labels shows Flat JSON input gives LLMs near-perfect record extraction F1 (0.9567) with low hallucination, far outperforming Slimmed HTML and Hierarchical JSON.
A list of websites is in the appendix, and prompting strategies are described
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.DB 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
NEXT-EVAL: Next Evaluation of Traditional and LLM Web Data Record Extraction
A new evaluation framework with XPath-based labels shows Flat JSON input gives LLMs near-perfect record extraction F1 (0.9567) with low hallucination, far outperforming Slimmed HTML and Hierarchical JSON.