TY - GEN
T1 - Hiding in Plain Sight
T2 - 2023 Findings of the Association for Computational Linguistics: EMNLP 2023
AU - Cooper, Portia
AU - Surdeanu, Mihai
AU - Blanco, Eduardo
N1 - Publisher Copyright:
© 2023 Association for Computational Linguistics.
PY - 2023
Y1 - 2023
N2 - To avoid detection by current NLP monitoring applications, progenitors of hate speech often replace one or more letters in offensive words with homoglyphs, visually similar Unicode characters. Harvesting real-world hate speech containing homoglyphs is challenging due to the vast replacement possibilities. We developed a character substitution scraping method and assembled the Offensive Tweets with Homoglyphs (OTH) Dataset (N=90,788) with more than 1.5 million occurrences of 1,281 non-Latin characters (emojis excluded). In an annotated sample (n=700), 40.14% of the tweets were found to contain hate speech. We assessed the performance of seven transformer-based hate speech detection models and found that they performed poorly in a zero-shot setting (F1 scores between 0.04 and 0.52), but normalizing the data dramatically improved detection (F1 scores between 0.59 and 0.71). Training the models using the annotated data further boosted performance (highest micro-averaged F1 score=0.88, using five-fold cross validation). This study indicates that a dataset containing homoglyphs known and unknown to the scraping script can be collected, and that neural models can be trained to recognize camouflaged real-world hate speech.
AB - To avoid detection by current NLP monitoring applications, progenitors of hate speech often replace one or more letters in offensive words with homoglyphs, visually similar Unicode characters. Harvesting real-world hate speech containing homoglyphs is challenging due to the vast replacement possibilities. We developed a character substitution scraping method and assembled the Offensive Tweets with Homoglyphs (OTH) Dataset (N=90,788) with more than 1.5 million occurrences of 1,281 non-Latin characters (emojis excluded). In an annotated sample (n=700), 40.14% of the tweets were found to contain hate speech. We assessed the performance of seven transformer-based hate speech detection models and found that they performed poorly in a zero-shot setting (F1 scores between 0.04 and 0.52), but normalizing the data dramatically improved detection (F1 scores between 0.59 and 0.71). Training the models using the annotated data further boosted performance (highest micro-averaged F1 score=0.88, using five-fold cross validation). This study indicates that a dataset containing homoglyphs known and unknown to the scraping script can be collected, and that neural models can be trained to recognize camouflaged real-world hate speech.
UR - http://www.scopus.com/inward/record.url?scp=85183305347&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85183305347&partnerID=8YFLogxK
M3 - Conference contribution
AN - SCOPUS:85183305347
T3 - Findings of the Association for Computational Linguistics: EMNLP 2023
SP - 2922
EP - 2929
BT - Findings of the Association for Computational Linguistics
PB - Association for Computational Linguistics (ACL)
Y2 - 6 December 2023 through 10 December 2023
ER -