Albanian Fact-Checked News Corpus

Albanian Fact-Checked News Corpus is a dataset of 809 Albanian news items whose veracity was checked by Faktoje.al, the Albanian fact-checking portal. It accompanies the article “Beyond Semantics: Structural, Syntactic and Topical Signals for Misinformation Detection in a Low-Resource Language, an Albanian Case” (Muraku and Kajo Meçe). Corpus (dataset_manually_reviewed_2026-10-30_v3.csv, UTF-8): 809 items, 423 labelled false and 386 true. Columns: source_link: the Faktoje fact-check page; original_link: the source of the checked claim; title; text: the article body, or a transcript when the claim came from a video; full_text: title + text; is_fake: 1 = false, 0 = true, according to Faktoje’s verdict. Full articles and short claims: 783 items are full articles (403 false, 380 true). The other 26 are short claims under 100 characters (20 false, 6 true), mostly social-media posts; they are kept in the file but were not analysed in the article. Features (features_structural_pos_recompute_v5.csv): 42 features per article, computed on the article body: structural and part-of-speech features; lexical richness (MTLD, MATTR, hapax ratio, lexical density); Albanian grammatical features (person, tense, passive voice, future and conditional forms); verifiability features (numbers, percentages, currency, dates). Collection and annotation Articles were collected from Faktoje.al with a crawler that followed the site’s robots.txt and waited between requests. Video claims were transcribed with youtube_transcript_api or Whisper (Albanian). Labels follow Faktoje’s published verdicts. If Faktoje verdict was E VERTETE (True) the assigned label is 0, otherwishe if the verdict was: E PAVERTETE/GENJESHTER (not true, lie) the assigned label is 1. Records were reviewed semi-manually. Items with empty text, incorrect transcripts, broken links or non-Albanian text were removed. Date range: 2018-2026. Access conditions The article texts and transcripts remain under the copyright of their original publishers and of Faktoje, and false articles name real people. The files are therefore shared on request, for non-commercial research only. To request access, use the “Request access” button and state your name, institution and intended research use. By requesting access you agree: not to redistribute the texts; not to use the data to identify, profile or target the individuals named in it; to cite this record and the accompanying article.Computer and information sciences License: labels and features under CC BY-NC 4.0. The article texts are not licensed by the authors.