Institutional-QA-PT: A Portuguese Question Answering Dataset for Institutional Information

Institutional-QA-PT is a Portuguese-language question answering dataset about the Computer Science graduate program at the Federal University of São Paulo (UNIFESP). The file corpus_completo.json contains 514 original questions and 1,507 non-empty paraphrases, with reference contexts, answer texts, and character offsets for extractive question answering. The dataset accompanies the study “Local Institutional QA under Resource and Privacy Constraints: Benchmarking Extractive, Generative, and Verified Hybrid Pipelines in Portuguese”. Privacy review: individual email addresses and personal profile links were replaced, and the student representative’s name was pseudonymized. Public faculty names, general institutional contact details, and links to institutional student and graduate listings remain. The dataset is not fully anonymous. This record contains the QA corpus. Source documents, split definitions, code, and evaluation artifacts are available in the associated repository: https://github.com/GPAM-ai/Institutional-QA-PT The study uses a validation holdout rather than an independent test set.