RECORD
Character-level HyperNetworks for Hate Speech Detection
Computation and Language · Submitted 11 Nov 2021
abstract compiled · 2026-09-23 · no language-model review
The authors' main claim Our results show that the proposed HyperNetworks achieve performance that is competitive, and better in some cases, than these pretrained language models, while being smaller by orders of magnitude.
The compiled record
The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.
Open the compiled record
The abstract, sentence by sentence
The massive spread of hate speech, hateful content targeted at specific subpopulations, is a problem of critical social 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, adapting these models to the task of hate speech detection using relevant labeled datasets. Unfortunately, there are only a few public labeled datasets of limited size that are available for this purpose. We make several contributions with high potential for advancing this state of affairs. We present HyperNetworks for hate speech detection, 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 orders of magnitude smaller compared to the popular DL classifiers. We further show that training hate detection classifiers using additional large amounts of automatically generated examples is beneficial in general, yet this practice especially boosts the performance of the proposed HyperNetworks. We report the results of extensive experiments, assessing the performance of multiple neural architectures on hate detection using five public 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. In addition 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 proposed HyperNetworks achieve performance that is competitive, and better in some cases, than these pretrained language models, while being smaller by orders of magnitude.
background contribution method result interpretation main claim
| # | sentence | role | the authors' own work | states the main conclusion | strength |
|---|---|---|---|---|---|
| s0 | The massive spread of hate speech, hateful content targeted at specific subpopulations, is a problem of critical social importance. | background 1.00 | 0.04 | 0.02 | asserted 1.00 |
| s1 | 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, adapting these models to the task of hate speech detection using relevant labeled datasets. | background 1.00 | 0.06 | 0.03 | asserted 1.00 |
| s2 | Unfortunately, there are only a few public labeled datasets of limited size that are available for this purpose. | gap 0.86 | 0.09 | 0.02 | asserted 1.00 |
| s3 | We make several contributions with high potential for advancing this state of affairs. | contribution 0.98 | 0.33 | 0.03 | hedged possibility 0.74 |
| s4 | We present HyperNetworks for hate speech detection, a special class of DL networks whose weights are regulated by a small-scale auxiliary network. | contribution 0.94 | 0.90 | 0.06 | asserted 1.00 |
| s5 | These architectures operate at character-level, as opposed to word or subword-level, and are several orders of magnitude smaller compared to the popular DL classifiers. | method 0.72 | 0.58 | 0.09 | asserted 1.00 |
| s6 | We further show that training hate detection classifiers using additional large amounts of automatically generated examples is beneficial in general, yet this practice especially boosts the performance of the proposed HyperNetworks. | result 0.78 | 0.93 | 0.12 | asserted 1.00 |
| s7 | We report the results of extensive experiments, assessing the performance of multiple neural architectures on hate detection using five public datasets. | method 0.65 | 0.95 | 0.04 | asserted 0.96 |
| s8 | 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. | method 0.95 | 0.58 | 0.02 | asserted 0.96 |
| s9 | In addition 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. | method 0.85 | 0.95 | 0.03 | asserted 0.92 |
| s10 | Our results show that the proposed HyperNetworks achieve performance that is competitive, and better in some cases, than these pretrained language models, while being smaller by orders of magnitude. | result 0.98 | 0.97 | 0.85 | asserted 1.00 |
What was examined, and what was not
- arXiv listing abstract v2
- full text
- figures and tables
- references and citing sentences
- relations to other papers
- P abstract
- S sections
- T body sentences
- C citations
- R cross-paper relations
The paper
Original abstract
The massive spread of hate speech, hateful content targeted at specific subpopulations, is a problem of critical social 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, adapting these models to the task of hate speech detection using relevant labeled datasets. Unfortunately, there are only a few public labeled datasets of limited size that are available for this purpose. We make several contributions with high potential for advancing this state of affairs. We present HyperNetworks for hate speech detection, 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 orders of magnitude smaller compared to the popular DL classifiers. We further show that training hate detection classifiers using additional large amounts of automatically generated examples is beneficial in general, yet this practice especially boosts the performance of the proposed HyperNetworks. We report the results of extensive experiments, assessing the performance of multiple neural architectures on hate detection using five public 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. In addition 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 proposed HyperNetworks achieve performance that is competitive, and better in some cases, than these pretrained language models, while being smaller by orders of magnitude.
Human review
How the literature uses it
What this paper does with the arXiv papers it cites
What it cites · 4 sentences without a confident label
Compares with 1 paper
-
We assess the proposed models alongside several deep learning networks, including the popular large transformer-based BERT and RoBERTA [ref] architectures, which model text at wordpiece level.
RoBERTa: A Robustly Optimized BERT Pretraining Approach · Introduction · 0.90
Discussion
Know something about this paper? Leave a comment, or highlight any passage above to comment on it.