Publish or Hold? Automatic Comment Moderation in Luxembourgish News Articles

Proceedings of Recent Advances in Natural Language Processing, pages 968–978 Varna, Sep 4–6, 2023 https://doi.org/10.26615/978-954-452-092-2_104 968 Publish or Hold? Automatic Comment Moderation in Luxembourgish News Articles Tharindu Ranasinghe♡, Alistair Plum♢, Christoph Purschke♢, Marcos Zampieri♠ ♡Aston University, Birmingham, UK ♢University of Luxembourg, Esch-sur-Alzette, Luxembourg ♠George Mason University, Fairfax, VA, USA t.ranasinghe@aston.ac.uk, alistair.plum@uni.lu christoph.purschke@uni.lu, mzampier@gmu.edu Abstract Recently, the internet has emerged as the pri- mary platform for accessing news. In the ma- jority of these news platforms, the users now have the ability to post comments on news ar- ticles and engage in discussions on various social media. While these features promote healthy conversations among users, they also serve as a breeding ground for spreading fake news, toxic content, and hate speech. Moderat- ing or removing such content is paramount to avoid unwanted consequences for the readers. However, apart from a few notable exceptions, most research on the automatic moderation of news article comments has dealt with English and other high-resource languages. This leaves under-represented or low-resource languages at a loss. Addressing this gap, we perform the first large-scale qualitative analysis of more than one million Luxembourgish comments posted over the course of 14 years. We eval- uate the performance of state-of-the-art trans- former models in Luxembourgish news article comment moderation. Furthermore, we anal- yse how the language of Luxembourgish news article comments has changed over time. We observe that machine learning models trained on old comments do not perform well on recent data. The findings in this work will be bene- ficial in building news comment moderation systems for many low-resource languages. 1 Introduction In recent years, the Internet has revolutionised how individuals access and consume news. With the popularity of smart devices such as phones and tablets, the Internet has emerged as the primary medium for acquiring news and information (Kwak et al., 2010). People often share news articles on social media using these devices and discuss them with their friends. At the same time, news web- sites also allow users to post comments and discuss stories (Zannettou et al., 2017). While the inclusion of comment sections pro- vides users with a platform to engage in construc- tive discussions regarding news stories, these dis- cussions can also devolve into the expression of offensive remarks and hate speech (Erjavec and Kovaˇciˇc, 2012; Davidson et al., 2017; Chowdhury et al., 2020). Furthermore, malicious users can ex- ploit discussion platforms to intentionally spread misinformation, often in the form of fake news, to mislead and provoke readers (Risch and Kres- tel, 2018; Yanagi et al., 2020). The wide spread of inappropriate comments motivates the use of content moderation to avoid further undesirable consequences. Moderating comment sections is a difficult task, mainly due to how widely the content can range, including fake news (Patwa et al., 2021) and vari- ous forms of offensive speech (Risch and Krestel, 2018; Napoles et al., 2017; Zampieri et al., 2019a; Weerasooriya et al., 2023). Detecting these varied types of content is difficult for humans alone, and in addition, the …

Content truncated to 500 words