Sagi Pendzel

Sagi Pendzel, Tomer Wullach, Amir Adler et al.

Automatic hate speech detection using deep neural models is hampered by the scarcity of labeled datasets, leading to poor generalization. To mitigate this problem, generative AI has been utilized to generate large amounts of synthetic hate speech sequences from available labeled examples, leveraging the generated data in finetuning large pre-trained language models (LLMs). In this chapter, we provide a review of relevant methods, experimental setups and evaluation of this approach. In addition to general LLMs, such as BERT, RoBERTa and ALBERT, we apply and evaluate the impact of train set augmentation with generated data using LLMs that have been already adapted for hate detection, including RoBERTa-Toxicity, HateBERT, HateXplain, ToxDect, and ToxiGen. An empirical study corroborates our previous findings, showing that this approach improves hate speech generalization, boosting recall performance across data distributions. In addition, we explore and compare the performance of the finetuned LLMs with zero-shot hate detection using a GPT-3.5 model. Our results demonstrate that while better generalization is achieved using the GPT-3.5 model, it achieves mediocre recall and low precision on most datasets. It is an open question whether the sensitivity of models such as GPT-3.5, and onward, can be improved using similar techniques of text generation.

Sagi Pendzel, Nir Lotan, Alon Zoizner et al.

The rise of social media has been argued to intensify uncivil and hostile online political discourse. Yet, to date, there is a lack of clarity on what incivility means in the political sphere. In this work, we utilize a multidimensional perspective of political incivility, developed in the fields of political science and communication, that differentiates between impoliteness and political intolerance. We present state-of-the-art incivility detection results using a large dataset of 13K political tweets, collected and annotated per this distinction. Applying political incivility detection at large-scale, we observe that political incivility demonstrates a highly skewed distribution over users, and examine social factors that correlate with incivility at subpopulation and user-level. Finally, we propose an approach for modeling social context information about the tweet author alongside the tweet content, showing that this leads to improved performance on the task of political incivility detection. We believe that this latter result holds promise for socially-informed text processing in general.