Offensive Language Detection on Video Live Streaming Chat

Proceedings of the 28th International Conference on Computational Linguistics, pages 1936–1940 Barcelona, Spain (Online), December 8-13, 2020 1936 Offensive Language Detection on Video Live Streaming Chat Zhiwei Gao, Shuntaro Yada, Shoko Wakamiya, and Eiji Aramaki Nara Institute of Science and Technology, Japan {gao.zhiwei.fw1,s-yada,wakamiya,aramaki}@is.naist.jp Abstract This paper presents a prototype of a chat room that detects offensive expressions in a video live streaming chat in real time. Focusing on Twitch, one of the most popular live streaming platforms, we created a dataset for the task of detecting offensive expressions. We collected 2,000 chat posts across four popular game titles with genre diversity (e.g., competitive, violent, peaceful). To make use of the similarity in offensive expressions among different social media platforms, we adopted state-of-the-art models trained on offensive expressions from Twitter for our Twitch data (i.e., transfer learning). We investigated two similarity measurements to predict the transferability, textual similarity, and game-genre similarity. Our results show that the transfer of features from social media to live streaming is effective. However, the two measurements show less correlation in the transferability prediction. 1 Introduction Posting offensive comments is a common problem of abusive behavior not only on representative social media platforms (e.g., Twitter) but also on other growing platforms1. For example, offensive expressions in live streaming chat could appear more frequently because users communicate with others in real time with less introspection during live streaming. In our preliminary study using crowdsourcing, seven out of 100 users of live streaming platforms answered that they had sent offensive expressions, while ten users answered that they had been the target of offensive behaviors. Tackling this problem on social media can provide more protection to users and more inspection methods for service providers. Offensive language detection on Twitter has already attracted much attention due to Twitter’s large user base and linguistic idiosyncrasies. For example, SemEval-2019 Task 6: Identifying and Catego- rizing Offensive Language in Social Media (Zampieri et al., 2019b) was organized as a shared task for offensive language detection on Twitter. The task consisted of three sub-tasks: (a) offensive language identification, (b) categorization of offense types, and (c) offense target identification. To provide more diversified characteristics for research on offensive language detection, it is also crucial to focus on video live streaming platforms that share similar features to Twitter, such as Twitch2, which is a popular live streaming platform mainly for gamers. This paper introduces a chat room that detects offensive expressions in real time and filters them out, as illustrated in Figure 1. To detect offensive expressions, we apply a deep learning approach. Modern neural architectures for natural language processing are generally highly effective when provided with a large amount of labeled training data (Dai et al., 2019). However, large volumes of labeled data are difficult to obtain, especially for video live streaming, as it is an emerging platform. Therefore, we apply a transfer learning approach to solve this problem. Domain adaptation, a sub-field of transfer learning, aims to learn a transferable representation from a source domain …

Content truncated to 500 words