Hate Speech Detection in Social Media for the Kurdish Language

With the rapid growth of technology over the world, especially, on the
internet, people enormously use social media freely to express their ideologies.
Sometimes the freedom of media is caught up and the rights of others are beaten
down by hate speech. Moreover, social media is an easy and vague way to desecrate
people, groups, and parties since there is no any way to recognize anonymous
users over social media. Testing human speech is common for English, Arabic,
and Turkish languages while there is no attempt for the Kurdish language. For
that reason, the Kurdish hate speech dataset is collected from comments on the
Facebook application as an effort for detecting hate speech and removing them.
The raw dataset consists of 6882 comments which are divided into hate and hot
hate classes. Support Vector Machine (SVM), Decision Tree (DT), and Naïve
Bays (NB) algorithms are implemented and compared. Based on the results, the
SVM is found most excellent with the F1 measure being 0.687.

随着全球科技尤其是互联网技术的飞速发展，人们愈发频繁地通过社交媒体自由表达自身的思想理念。然而仇恨言论有时会践踏媒体自由，侵害他人权益。此外，由于社交媒体无法识别匿名用户，其便捷性与身份隐匿的特性，使其成为污蔑个人、群体与政党的便捷渠道。针对英语、阿拉伯语与土耳其语的仇恨言论检测研究已较为常见，但目前尚无针对库尔德语的相关尝试。为此，本研究从脸书（Facebook）平台的评论中收集了库尔德语仇恨言论数据集，以期为仇恨言论检测与清除工作提供支撑。该原始数据集共包含6882条评论，分为仇恨类与重度仇恨类两个类别。本研究实现了支持向量机（Support Vector Machine, SVM）、决策树（Decision Tree, DT）与朴素贝叶斯（Naive Bayes, NB）三种算法并开展对比实验。实验结果表明，支持向量机的表现最优，其F1分数达到0.687。