TuEval at SemEval-2019 Task 5: LSTM Approach to Hate Speech Detection in English and Spanish

Proceedings of the 13th International Workshop on Semantic Evaluation (SemEval-2019), pages 498–502 Minneapolis, Minnesota, USA, June 6–7, 2019. ©2019 Association for Computational Linguistics 498 TuEval at SemEval-2019 Task 5: LSTM Approach to Hate Speech Detection in English and Spanish Mihai Manolescu Denise L¨offlad Adham Nasser Mohamed Saber Masoumeh Moradipour Tari {mihai.manolescu, denise.loefflad, adham-nasser.mohamed-saber, masoumeh.moradipour-tari} @student.uni-tuebingen.de Eberhard-Karls University of T¨ubingen Abstract The detection of hate speech, especially in on- line platforms and forums, is quickly becom- ing a hot topic as anti-hate speech legislation begins to be applied to public discourse on- line. The HatEval shared task was created with this in mind; participants were expected to de- velop a model capable of determining whether or not input (in this case, Twitter posts in En- glish and Spanish) could be considered hate speech (designated as Subtask A), if they were aggressive, and whether the tweet was target- ing an individual, or speaking generally (Sub- task B). We approached this Subtask by creat- ing a LSTM model with an embedding layer. We found that our model performed consid- erably better on English language input when compared to Spanish language input. In En- glish, we achieved an F1-Score of 0.466 for Subtask A and 0.462 for Subtask B; In Span- ish, we achieved scores of 0.617 and 0.612 on Subtask A and Subtask B, respectively. 1 Introduction Social media plays an important role nowadays and dominates everyday life. Social networks like Facebook, Twitter and Instagram are platforms where people express thoughts, feelings and emotions regarding themselves or others. This can lead to different opinions colliding and creating conflicts. Often, feelings are not expressed objec- tively and can be offensive to other users. In order to make social media more comfortable, so called hate speech needs to be detected and removed. Hate speech is here defined as: Any communi- cation that disparages a person or a group on the basis of some characteristic such as race, color, ethnicity, gender, sexual orientation, nationality, religion, or other characteristics (Basile et al., 2019). To assure there is no spread of illegal hate speech, the EU has created a code of conduct for social media platforms (European Union, 2018) that needs to be followed. According to these EU regulations, social media platforms must regulate hateful speech. In addition, social media occupies an increasingly larger portion of public discourse; even without these EU regulations, it seems that these platforms should have some methods for controlling violent discourse. For these reasons, the HatEval shared task (Basile et al., 2019) was created. The task is divided into two subtasks; Subtask A is hate speech detection against immigrants and women, a binary classification problem where a tweet is classified as either hateful or not hateful. Subtask B is determining whether a given tweet is aggressive, and whether it is targeting an individual, or not referring to any particular person. Further, each of these Subtasks is evaluated on English tweets and using Spanish tweets. We were provided a 9000-tweet English training set, and …

Content truncated to 500 words