Cultural Topic Modelling over Novel Wikipedia Corpora for South-Slavic Languages

Proceedings of Recent Advances in Natural Language Processing, pages 910–917 Sep 1–3, 2021. https://doi.org/10.26615/978-954-452-072-4_104 910 Cultural Topic Modelling over Novel Wikipedia Corpora for South-Slavic Languages Filip Markoski1, Elena Markoska1, Nikola Ljubeˇsi´c2, Eftim Zdravevski1, and Ljupcho Kocarev3,1 1Faculty of Computer Science and Engineering, Ss. Cyril and Methodius University in Skopje, Macedonia, 1e-mails: filip.markoski45@gmail.com, elena.markoska@live.com, eftim.zdravevski@finki.ukim.mk 2Joˇzef Stefan Institute, Slovenia, e-mail: nikola.ljubesic@ijs.si 3Macedonian Academy of Sciences and Arts, Skopje, Macedonia, e-mail: lkocarev@manu.edu.mk Abstract There is a shortage of high-quality corpora for South-Slavic languages. Such corpora are useful to computer scientists and researchers in social sciences and humanities alike, fo- cusing on numerous linguistic, content anal- ysis, and natural language processing appli- cations. This paper presents a workflow for mining Wikipedia content and processing it into linguistically-processed corpora, applied on the Bosnian, Bulgarian, Croatian, Macedo- nian, Serbian, Serbo-Croatian and Slovenian Wikipedia. We make the resulting seven cor- pora publicly available. We showcase these corpora by comparing the content of the under- lying Wikipedias, our assumption being that the content of the Wikipedias reflects broadly the interests in various topics in these Balkan nations. We perform the content compari- son by using topic modelling algorithms and various distribution comparisons. The results show that all Wikipedias are topically rather similar, with all of them covering art, cul- ture, and literature, whereas they contain dif- ferences in geography, politics, history and sci- ence. 1 Introduction Researchers studying the South-Slavic languages often face difficulties finding corpora in the re- spective languages. While for Slovenian there is a significant amount of corpora available both for search and download (Krek et al., 2020; Fiˇser et al., 2020; Erjavec et al., 2020), most other lan- guages in this language group do not enjoy this commodity. This is the reason why web corpora with all their limitations and uncertainties are so popular in this language group1. This paper de- scribes an effort to add one additional, consistent 1The paper describing the Croatian National Corpus has been cited since 2017 on Google Scholar 19 times, while the paper describing the older version of the Croatian web corpus was cited in the same period 58 times. Some other source of good-quality text for South-Slavic lan- guages - Wikipedia corpora. To perform that, we set up a robust pipeline for preparing and linguisti- cally processing all currently available Wikipedias of South-Slavic (macro-)languages. We use the notion of (macro-)languages to signify that the se- lected Wikipedias refer to six national languages with a respective ISO-639-1 code, and one macro- language, Serbo-Croatian, with its ISO-639-3 code hbs. Processing Wikipedia corpora could seem a rela- tively simple task for researchers in computer sci- ence, and therefore such processing is performed on a project basis, as was already done for some South-Slavic languages (Ljubeˇsi´c and Fiˇser, 2013; Svoboda and Beliga, 2017). However, other scien- tific disciplines lack the technical expertise to per- form these data preparations. Furthermore, process- ing Wikipedia data on a per-project basis entirely disregards the questions of replicability and repro- ducibility of research, making the …

Content truncated to 500 words