Finnish Hate-Speech Detection on Social Media Using CNN and FinBERT

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 876–882 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 876 Finnish Hate-Speech Detection on Social Media Using CNN and FinBERT Md Saroar Jahan, Mourad Oussalah, Nabil Arhab University Of Oulu CMVS, BP 4500, 90014 Finland {Md.Jahan, Mourad.Oussalah,nabil.arhab}@oulu.fi Abstract There has been a lot of research in identifying hate posts from social media because of their detrimental effects on both individuals and society. The majority of this research has concentrated on English, although one notices the emergence of multilingual detection tools such as multilingual-BERT (mBERT). However, there is a lack of hate speech datasets compared to English, and a multilingual pre-trained model often contains fewer tokens for other languages. This paper attempts to contribute to hate speech identification in Finnish by constructing a new hate speech dataset that is collected from a popular forum (Suomi24). Furthermore, we have experimented with FinBERT pre-trained model performance for Finnish hate speech detection compared to state-of-the-art mBERT and other practices. In addition, we tested the performance of FinBERT compared to fastText as embedding, which employed with Convolution Neural Network (CNN). Our results showed that FinBERT yields a 91.7% accuracy and 90.8% F1 score value, which outperforms all state-of-art models, including multilingual-BERT and CNN. Keywords: Hate detection, Finnish hate, FinBERT, Finnish Hate dataset 1. Introduction The proliferation of online social media platforms with mil- lions of user-generated content every day has witnessed a substantial increase of online hate speech (HS), raising the concern of society, individuals, policymakers, and re- searchers. (Brown, 2017) has defined hate speech as any textual or verbal practice that implicates issues of discrimination or violence against people regarding their race, ethnicity, na- tionality, religion, sexual orientation, and gender identity. According to (Anis and Maret, 2017) hate speech can occur in different linguistic styles and several acts like insulting, provocation, abusing, and aggression. However, according to (Chetty and Alathur, 2018), hate speech can be catego- rized into gendered, religious, and racist hate speech. Especially, offensive language promotes discrimination based on gender, nationality, or race (Nockleby, 2000), which often leads to profound adverse effects, particularly for teenage groups, organizations, and society. Many in- ternet companies offer criteria and generic guidelines that users must adhere to when posting content on certain sites to solve this problem. Simultaneously, they use manual an- notators to detect offensive language and remove it accord- ingly. However, such a manual effort is invariably expen- sive, non-scalable, and non-sustainable, inspiring the bene- fit of a solution based on an automatic offensive language detection. Prior work has studied detection of offensive language spe- cially for English language in Twitter (Foong and Oussalah, 2017; Wiegand et al., 2018), FromSpring posts (Reynolds et al., 2011), Wikipedia comments and Facebook posts (Ku- mar et al., 2018). Besides, scholars examined various char- acteristics of offensive language such as the cyber aggres- sion (Kumar et al., 2018), hate speech (Abderrouaf and Oussalah, 2019), abusive language (Mubarak et al., 2017), …

Content truncated to 500 words