pith. sign in

arxiv: 2205.00501 · v1 · pith:HFMRZ7QN · submitted 2022-05-01 · cs.HC · cs.AI· cs.CL· cs.LG

Is Your Toxicity My Toxicity? Exploring the Impact of Rater Identity on Toxicity Annotation

pith:HFMRZ7QNopen to challenge →

classification cs.HC cs.AIcs.CLcs.LG
keywords raterpoolsraterstoxicitycommentsmodelsannotationsidentity
0
0 comments X
read the original abstract

Machine learning models are commonly used to detect toxicity in online conversations. These models are trained on datasets annotated by human raters. We explore how raters' self-described identities impact how they annotate toxicity in online comments. We first define the concept of specialized rater pools: rater pools formed based on raters' self-described identities, rather than at random. We formed three such rater pools for this study--specialized rater pools of raters from the U.S. who identify as African American, LGBTQ, and those who identify as neither. Each of these rater pools annotated the same set of comments, which contains many references to these identity groups. We found that rater identity is a statistically significant factor in how raters will annotate toxicity for identity-related annotations. Using preliminary content analysis, we examined the comments with the most disagreement between rater pools and found nuanced differences in the toxicity annotations. Next, we trained models on the annotations from each of the different rater pools, and compared the scores of these models on comments from several test sets. Finally, we discuss how using raters that self-identify with the subjects of comments can create more inclusive machine learning models, and provide more nuanced ratings than those by random raters.

This paper has not been read by Pith yet.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. IYKYK (But AI Doesn't): Automated Content Moderation Does Not Capture Communities' Heterogeneous Attitudes Towards Reclaimed Language

    cs.CL 2026-04 unverdicted novelty 5.0

    Automated hate speech detectors show poor alignment with heterogeneous in-group judgments on reclaimed slur usage, driven by low inter-annotator agreement and contextual features like derogatory intent.

  2. PaLM 2 Technical Report

    cs.CL 2023-05 unverdicted novelty 5.0

    PaLM 2 reports state-of-the-art results on language, reasoning, and multilingual tasks with improved efficiency over PaLM.

  3. Reducing the rate of personal insults in social media with bystander bots

    cs.SI 2026-06 unverdicted novelty 4.0

    A randomized controlled trial on Reddit found that automated deescalation replies, especially appreciation messages, reduced the rate of personal insults posted by users.