FakenewsBR: A Unified Portuguese Misinformation Corpus for Reproducible Data Integration Research

Abstract

The growing impact of misinformation in digital environments has intensified the demand for structured and reusable datasets to support empirical research. This paper presents FakenewsBR, a consolidated Portuguese-language fake news resource comprising 51,205 instances integrated from six previously curated corpora rather than newly collected raw data. Through a reproducible pipeline, heterogeneous sources were harmonized under a unified schema, subjected to controlled cleaning procedures, and processed using near-duplicate analysis to flag highly similar records for manual review. The final dataset preserves provenance information, spans a temporal window from 2005 to 2022, and includes fact-check metadata for 18.18% of instances. Rather than proposing new detection models, the contribution focuses on systematic integration, structural normalization, and transparent documentation to facilitate reuse in machine learning, database research, and misinformation analysis. The dataset and its construction pipeline are publicly available at: https://github.com/AKCIT-FN/fakenews-data.