Serbian CLASSLA-web: Corpus of the Serbian Web
The Serbian Web Corpus (CLASSLA-web.sr) is a Serbian corpus made up of texts collected from the Internet. The corpus belongs to the CLASSLA-web corpus collection. The corpora are built from national top-level domain web crawls and are structured, cleaned, and automatically enriched with linguistic, genre, and topic annotations.
For detailed information about the collection, see the CLASSLA-web project page.
The most recent version of the Serbian corpus (CLASSLA-web.sr 2.0) consists of 3 billion words. The texts were downloaded during a 2024 web crawl primarily from the .rs and .срб domains. High-frequency web domains were manually inspected, and automatically generated or poor-quality content was removed. To ensure better quality, this corpus is not an exact copy of the original dataset. We manually inspected the most frequent websites and completely removed all documents from entire sites that contained spam or low-quality content, such as lists, tables, and unfinished sentences. For the 2024 release (sr 24), we removed 15 bad websites totaling 166,810 documents (2.3% of the original document count). In the earlier 2022 release (sr 22), 4 bad websites accounting for 25,021 documents (0.5% of the original count) were also removed.
Acknowledgments: The corpus was developed and curated by the CLASSLA knowledge centre and distributed via the CLARIN.SI repository. We acknowledge the authors of the CLASSLA-web collection—Taja Kuzman Pungeršek, Peter Rupnik, Vít Suchomel, and Nikola Ljubešić—for their work in creating these resources.
Part-of-speech tagset and lemmatization
The corpus was tagged with MULTEXT-East Croatian part-of-speech tagset version 5 and lemmatized with cstlemma.
Overview of Serbian Classla corpora
These web corpora were crawled and processed repeatedly during the years:
- Serbian CLASSLA-web 2.0 (2024) (classla_sr24_rft1) – 3 billion words, genre annotation and topic classification
- Serbian CLASSLA-web 1.0 (2022) (classla_sr22_rft1) – 2.3 billion words, genre annotation
Serbian CLASSLA-web 2024 corpus sizes
| Number of words | 3+ billion |
| Number of tokens | 3.6+ billion |
| Number of sentences | 184+ million |
| Number of web pages | 7+ million |
Search the Serbian CLASSLA-web 2024
Sketch Engine offers a range of tools to work with this Serbian corpus.
Tools to work with the Serbian CLASSLA-web corpus from the web
A complete set of Sketch Engine tools is available to work with this Serbian corpus to generate:
- word sketch – Serbian collocations categorized by grammatical relations
- thesaurus – synonyms and similar words for every word
- keywords – terminology extraction of one-word and multi-word units
- word lists – lists of Serbian nouns, verbs, adjectives etc. organized by frequency
- n-grams – frequency list of multi-word units
- concordance – examples in context
- text type analysis – statistics of metadata in the corpus
Changelog
Serbian CLASSLA-web
version 2.0 (2024)
- published in Sketch Engine in 10/2026
- further spam and low-quality text removal
version 1.0 (2022)
- published in Sketch Engine in 10/2026
- further spam and low-quality text removal
Bibliography
Ljubešić, Nikola, and Taja Kuzman. “CLASSLA-web: Comparable Web Corpora of South Slavic Languages Enriched with Linguistic and Genre Annotation.” Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024). 2024.
Use Sketch Engine in minutes
Generate collocations, frequency lists, examples in contexts, n-grams or extract terms. Use our Quick Start Guide to learn it in minutes.
