Corpus InfOpinion-RTBF

Bogaert, J., Escouflaire, L., de Marneffe, M. C., Descampe, A., Standaert, F. X., & Fairon, C. (2023). TIPECS: A corpus cleaning method using machine learning and qualitative analysis. In International Conference on Corpus Linguistics (JLC). Please read the "Terms of use" file before using the corpus. Values in the CSV file are separated by the character "|". The InfOpinion corpus consists of 10,000 articles published between 2012 and 2021 on the infor- mation feed of the RTBF Corpus. It is a balanced dataset containing on the one hand 5,000 articles classified by the RTBF as Op-eds (“Chroniques”) or Opinions, and on the other hand 5,000 articles randomly selected from the Belgium, World, and Society categories, which tackle similar topics to those discussed in the Op-eds category. Each article in the InfOpinion corpus has multiple layers of metadata, which are described further in (Escouflaire et al., 2024) : ID, title, publication date, signature, feed, category and keyword. An additional column is devoted to the article’s genre (or class), which is either information or opinion. This corpus was built for training and evaluating a classification model distinguishing between pieces from the journalistic opinion genre (such as editorials, commentaries, reviews), which are considered to be subjective, and pieces belonging to the information genre (press agency dispatches, news articles), meant to be more objective texts (Grosse, 2001). This binary categorization relies solely on the articles’ annotation by the RTBF as either opinion or information.