.png){kind=icon}
Return
Character-level HyperNetworks for Hate Speech Detection
.png){kind=icon}
OA
AI
DOI：10.1016/j.eswa.2022.117571.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The massive spread ofhate speech, hateful content targeted at specific subpopulations, is a problem of criticalsocial importance. Automated methods of hate speech detection typically employ state-of-the-art deep learning(DL)-based text classifiers-large pretrained neural language models of over 100 million parameters, adaptingthese models to the task of hate speech detection using relevant labeled datasets. Unfortunately, there areonly a few public labeled datasets of limited size that are available for this purpose. We make severalcontributions with high potential for advancing this state of affairs. We present HyperNetworks for hate speechdetection, a special class of DL networks whose weights are regulated by a small-scale auxiliary network.These architectures operate at character-level, as opposed to word or subword-level, and are several ordersof magnitude smaller compared to the popular DL classifiers. We further show that training hate detectionclassifiers using additional large amounts of automatically generated examples is beneficial in general, yetthis practice especially boosts the performance of the proposed HyperNetworks. We report the results ofextensive experiments, assessing the performance of multiple neural architectures on hate detection using fivepublic datasets. The assessed methods include the pretrained language models of BERT, RoBERTa, ALBERT,MobileBERT and CharBERT, a variant of BERT that incorporates character alongside subword embeddings. Inaddition to the traditional setup of within-dataset evaluation, we perform cross-dataset evaluation experiments,testing the generalization of the various models in conditions of data shift. Our results show that the proposedHyperNetworks achieve performance that is competitive, and better in some cases, than these pretrainedlanguage models, while being smaller by orders of magnitude
Keywords:
Hate speech detection
Neural networks
Text generation
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
7.5
Papers:
3.0W
Citations:
10.2W
Organization
Cited Papers
Rimegepant 75 mg for the Acute Treatment of Migraine in Adults With Frequent Migraine: Long-Term Safety and Clinical Improvement Versus Baseline (5054).png)
Neurology
NeurologyIF0
PREAI
ShareSaveRepeated and Time-Correlated Morphological Convergence in Cave-Dwelling Harvestmen (Opiliones, Laniatores) from Montane Western North AmericaPLoS ONEOAAI
PLoS ONEIF0
OAAIShareSave
ShareSaveShareSaveShareSave
View more