KDEHatEval at SemEval-2019 Task 5: A Neural Network Model for Detecting Hate Speech in Twitter

Proceedings of the 13th International Workshop on Semantic Evaluation (SemEval-2019), pages 365–370 Minneapolis, Minnesota, USA, June 6–7, 2019. ©2019 Association for Computational Linguistics 365 KDEHatEval at SemEval-2019 Task 5: A Neural Network Model for Detecting Hate Speech in Twitter Umme Aymun Siddiqua, Abu Nowshed Chy, and Masaki Aono Department of Computer Science and Engineering Toyohashi University of Technology, Toyohashi, Aichi, Japan. {aymun,nowshed}@kde.cs.tut.ac.jp and aono@tut.jp Abstract In the age of emerging volume of microblog platforms, especially twitter, hate speech prop- agation is now of great concern. However, due to the brevity of tweets and informal user gen- erated contents, detecting and analyzing hate speech on twitter is a formidable task. In this paper, we present our approach for detecting hate speech in tweets defined in the SemEval- 2019 Task 5. Our team KDEHatEval em- ploys different neural network models includ- ing multi-kernel convolution (MKC), nested LSTMs (NLSTMs), and multi-layer percep- tron (MLP) in a unified architecture. More- over, we utilize the state-of-the-art pre-trained sentence embedding models including Deep- Moji, InferSent, and BERT for effective tweet representation. We analyze the performance of our method and demonstrate the contribution of each component of our architecture. 1 Introduction Nowadays, microblog platforms such as twit- ter has become the most popular communica- tion medium among the people due to its conve- nient feature for sharing views, opinions, breaking news, and ideas. Besides its robust communica- tion feature, it has facilitated the evil-minded peo- ple to propagate anti-social behavior including on- line harassment, cyber-bullying, and hate speech. Hate speech is commonly defined as any com- munication that disparages a person or a group on the basis of some characteristics such as race, color, ethnicity, gender, sexual orientation, na- tionality, religion, or other characteristics (Basile et al., 2019). Given the massive amount of user- generated contents on the microblog, the problem of detecting, and therefore possibly limit the hate speech diffusion, is becoming fundamental, for in- stance for fighting against misogyny and xenopho- bia. To address the challenges of hate speech detec- tion in microblog platforms, Basile et al. (2019) proposed a multilingual detection of hate speech (HatEval) in twitter, task 5 at SemEval-2019. The task features two specific different targets includ- ing immigrants and women and focuses on two re- lated subtasks. Task A defines a two-class (or binary) classi- fication problem where a system needs to predict whether a tweet in English or Spanish with a given target (women or immigrants) is hateful or not hateful. Whereas task B defines the aggressive be- havior and target classification problem. A system first classifies a hateful tweet as aggressive or not aggressive, and then identify the target harassed as the individual or generic (i.e., single human or group). In this paper, we only focus on the English tweets for both task A and B. The rest of the paper is structured as follows: Section 2 provides a brief overview of prior re- search. In Section 3, we introduce our proposed neural network model. Section 4 includes exper- …

Content truncated to 500 words