KDELAB at SemEval-2020 Task 12: A System for Estimating Aggression of Tweets Using Two Layers of BERT Features

Proceedings of the 14th International Workshop on Semantic Evaluation, pages 2030–2034 Barcelona, Spain (Online), December 12, 2020. 2030 KDELAB at SemEval-2020 Task 12: A System for Estimating Aggression of Tweets Using Two layers of BERT Features Keisuke Hanahata, and Masaki Aono Department of Computer Science and Engineering Toyohashi University of Technology, Toyohashi, Aichi, Japan. hanahata@kde.cs.tut.ac.jp and aono@tut.jp Abstract In recent years, with the development of social network services and video distribution services, there has been a sharp increase in offensive posts. In this paper, we present our approach for detecting hate speech in tweets defined in the SemEval- 2020 Task 12. Specifically, we describe our ensemble system, based on “BERT Large”, having two parallel neural net component models, where one component model consists of MLP with input taken from the 23rd layer output from BERT, while another component model consists of bidirectional LSTM with input taken from the 1st layer of BERT. 1 Introduction The number of offensive posts has been increasing rapidly due to the wide spread of social networking services such as Facebook, Twitter, and YouTube, where obvious and/or latent offensive texts are ac- companied. It is necessary to filter the offensive postings because they can offend users and reduce their satisfaction with the service. Manual filtering is expensive, laborious, and time-consuming. Thus, the demand for automatic filtering has been increasing year by year. Marcos et al. presented SemEval2020 Task12 (Zampieri et al., 2020), which includes three sub-tasks. Sub-task A is an estimate of the aggressiveness of a tweet, Sub-taskB is an estimate of whether a tweet is attacked, and Sub-task C is an estimate of the target of a tweet’s attack. 2 Related Work A number of studies have been conducted to estimate the aggressiveness of documents. One study classifies Facebook and Twitter posts into three classes: explicitly offensive, covertly offensive, and non- offensive (Kumar et al., 2018), and another identifies hate speech. At Kaggle, a competition was held to classify offensive comments into six classes: toxic, more toxic, obscene, threatening, insulting, and personal attack. In Task 6 of SemEval2019 (Zampieri et al., 2019b), a competition was held for Twitter posts using a dataset with hierarchical labeling focusing on the target of the attack. 3 Proposed Method In the following section, we describe our proposed model for estimating the aggressiveness and the targets of tweets. 3.1 Preprocessing We perform preprocess, given all incoming tweets. The atmark, hash, and “URL” tokens associated with user IDs are removed because they are considered to interfere with the conversion of tweets into features. The “USER” token is left out because it can be a clue to estimate the target of the attack. Many of the tweets contain emojis. Emojis can contain information about the attacker or the target of an attack. The emoji library is used to convert emojis into text. This work is licensed under a Creative Commons Attribution 4.0 International License. License details: http:// creativecommons.org/licenses/by/4.0/. 2031 3.2 Feature extraction using BERT We think that information related to word aggressiveness …

Content truncated to 500 words