Proactively Reducing the Hate Intensity of Online Posts via Hate Speech Normalization

Abstract

Although pre-trained large language models (PLMs) have achieved state-of-the-art on many NLP tasks, they lack understanding of subtle expressions of implicit hate speech. Such nuanced and implicit hate is often misclassified as non-hate. Various attempts have been made to enhance the detection of (implicit) hate content by augmenting external context or enforcing label separation via distance-based metrics. We combine these two approaches and introduce FiADD, a novel Focused Inferential Adaptive Density Discrimination framework. FiADD enhances the PLM finetuning pipeline by bringing the surface form of an implicit hate speech closer to its implied form while increasing the inter-cluster distance among various class labels. We test FiADD on three implicit hate datasets and observe significant improvement in the two-way and three-way hate classification tasks. We further experiment on the generalizability of FiADD on three other tasks, namely detecting sarcasm, irony, and stance, in which surface and implied forms differ, and observe similar performance improvement. We analyze the generated latent space to understand its evolution under FiADD, which corroborates the advantage of employing FiADD for implicit hate speech detection. 1

Measuring Causal Effects of Civil Communication without Randomization

“…From a human-computer interaction perspective, research has focused on mitigating uncivil behaviors through platform-level intervention and policy (Chandrasekharan et al 2017;Jhaver et al 2021), encouraging user and community self-moderation (Seering, Kraut, and Dabbish 2017), and by examining how platform and UX design promote desirable undesirable behavior (Munn 2020;Seering et al 2019). Uncivil speech detection can be formulated as supervised machine learning and NLP tasks (Davidson et al 2017), often with the aim of building automated systems for filtering such unwanted content either tailored to specific platforms (Masud et al 2022;Nobata et al 2016) or for general use across contexts (Jigsaw 2021; Lees et al 2022). There is also a significant body of work on measuring and classifying uncivil behavior beyond textual communication in online gaming contexts (Canossa et al 2021;Kou 2020).…”

Revisiting Hate Speech Benchmarks: From Data Curation to System Deployment

“…Inspired by our experiments, in partnership with Wipro AI, we are developing an interactive web interface for contextual hate speech detection [34]. This interface will be a part of their more extensive pipeline to flag and analyze harmful content on the web.…”

“…• Benchmarking: We benchmark GOTHate with ten diverse and widely-studied baseline methods (Section 6). • Content moderation pipeline: This research has led to the creation of a hate speech detection pipeline currently under development in collaboration with Wipro AI [34,35] (Section 7). Reproducibility: The source code and sample dataset are publicly available on our Github 2 .…”