Material suplementar: Pipeline IFC → Classificação Semântica com LLMs: tabelas NIN, vocabulário controlado, arquivo IDS e dados de validação

This dataset contains the supplementary materials for the article "From IFC Metadata to Semantic Classification: Comparing Large Language Models under Controlled Information-Quality Scenarios" (Sofiste & Duarte, IME, 2026). It includes six files: (1) NIN_Pipeline_IFC_Suplementar.xlsx — Level of Information Need (LOI) tables for 13 IFC types, structured according to ISO 7817-1:2024; (2) Vocabulario_Item_EAP_Pipeline.xlsx — the controlled vocabulary of 71 valid values for the Item_EAP field of the Pipeline_Pset, mapped to NBR 15965 classification codes; (3) Pset_Pipeline_UserDefinedPsets_v3.txt — Autodesk Revit shared parameters file defining the 16 attributes of the Pipeline_Pset; (4) Pipeline_NIN.ids — buildingSMART Information Delivery Specification (IDS) file formalizing the NIN requirements for the 13 IFC types in scope; (5) validar_ids.py — Python script for validating IFC models against the Pipeline_NIN.ids file using the ifctester library; (6) Kappa_Validacao_Interavaliadores.xlsx — inter-rater reliability spreadsheet containing classifications by two independent evaluators, Cohen's Kappa calculations (κ = 0.9110), and precision/recall/F1-score metrics by category for each evaluated LLM.