Dataset on genre and news topic persistence between versions 1 and 2 of CLASSLA-web

  Authors

Ljubešić, Nikola ; Dinić, Vuk and Rupnik, Peter

  Item identifier

http://hdl.handle.net/11356/1972

 Project URL

https://clarinsi.github.io/classla-web/

 Date issued

2026-10-13

 Type

corpus, text

 Size

38057171 entries

 Language(s)

Bosnian , Bulgarian , Croatian , Macedonian , Montenegrin , Serbian , Slovenian

 Description

The dataset contains metadata on documents from the CLASSLA-web v2 web crawl (http://hdl.handle.net/11356/2079), describing each document by its (1) document identifier, (2) URL, (3) web domain, (4) predicted genre, (5) predicted topic, (6) the length in terms of number of characters and, most importantly, (7) whether the document is considered to have a near-duplicate in the CLASSLA-web v1 corpus collection. This dataset originated from the paper “What Persists across Web Crawls? Genre and News Topic Variation in South Slavic Web Corpora”, accepted for publication at the 13th Web-as-Corpus Workshop (WaC-13), co-located with EMNLP 2026. We make the dataset openly available to facilitate follow-up research on this topic.

 Publisher

Jožef Stefan Institute

 Acknowledgement

Jožef Stefan Institute CLARIN "CLARIN.SI"

European Union GA 101198470 "Digital Europe Programme: LLMs4EU"

 Subject(s)

web corpus web crawling document persistence

 Collection(s)

CLARIN.SI data & tools

Show full item record