Tw-StAR at SemEval-2019 Task 5: N-gram embeddings for Hate Speech Detection in Multilingual Tweets

Proceedings of the 13th International Workshop on Semantic Evaluation (SemEval-2019), pages 503–507 Minneapolis, Minnesota, USA, June 6–7, 2019. ©2019 Association for Computational Linguistics 503 Tw-StAR at SemEval-2019 Task 5: N-gram embeddings for Hate Speech Detection in Multilingual Tweets Hala Mulki∗, Chedi Bechikh Ali∗∗, Hatem Haddad†§ and Ismail Babao˘glu∗ ∗Department of Computer Engineering, Selcuk University, Turkey ∗∗LISI Laboratory, INSAT, Carthage University, Tunisia †RIADI Laboratory, National School of Computer Sciences, University of Manouba, Tunisia §iCompass Consulting, Tunisia halamulki@selcuk.edu.tr,chedi.bechikh@gmail.com haddad.Hatem@gmail.com,ibabaoglu@selcuk.edu.tr Abstract In this paper, we describe our contribution in SemEval-2019: subtask A of task 5 “Multilin- gual detection of hate speech against immi- grants and women in Twitter (HatEval)”. We developed two hate speech detection model variants through Tw-StAR framework. While the first model adopted one-hot encoding n- grams to train an NB classifier, the second gen- erated and learned n-gram embeddings within a feedforward neural network. For both mod- els, specific terms, selected via MWT patterns, were tagged in the input data. With two fea- ture types employed, we could investigate the ability of n-gram embeddings to rival one-hot n-grams. Our results showed that in English, n-gram embeddings outperformed one-hot n- grams. However, representing Spanish tweets by one-hot n-grams yielded a slightly better performance compared to that of n-gram em- beddings. The official ranking indicated that Tw-StAR ranked 9th for English and 20th for Spanish. 1 Introduction Under the guise of free speech, social media sys- tems have been misused by some users who em- bed hatred, offensive, racist or negative stereo- typing contents within their shared posts. Unfor- tunately, online Hate Speech (HS) is spreading widely, forming a serious problem that can lead to actual hate crimes (Matsuda, 2018). Many coun- tries adopted laws prohibiting HS where people convicted of using HS can face large fines and even imprisonment. Although Twitter has its anti HS policy∗, the increasing size of the daily-shared tweets in addition to multilingualism and informal writing issues evoke the necessity for automatic HS detection in tweets. ∗support.twitter.com/articles/ 20175050 Hate speech detection problem has been ad- dressed as a machine learning classification task. Recent studies proposed multiple HS detection models with different characteristic in terms of features, classification algorithms and implemen- tation architectures. While some HS models em- ployed hand-crafted features generated by NLP tools and external semantic resources, other mod- els adopted text embedding features that are auto- matically learned from the corpus itself. Both fea- ture types were fed to train either traditional clas- sifiers such as Support Vector Machines (SVM), Naive Bayes (NB) and so forth, or more compli- cated deep learning-based classifiers such as Con- volutional Neural Network (CNN), Long Short- Term Memory (LSTM) and Recurrent Neural Net- work (RNN) (Schmidt and Wiegand, 2017). The variety of hand-crafted features enabled obtain- ing reliable performances. However, generating such features based on morphological NLP tools or semantic resources remains laborious. In con- trast, embedding features are easier to generate and can yield good HS classification results when used within deep learning architectures (Yuan et al., 2016). …

Content truncated to 500 words