According to the arXiv preprint, research on gender-related online content has typically been operationalized through harm-centered tasks. The paper notes that this approach collapses non-harmful gender-related content into the same category as unrelated content, obscuring the full spectrum of discourse. It also points out that existing datasets are concentrated in high-resource languages, leaving many regions under-represented.

The paper introduces GENDIS, a dataset built from multilingual YouTube comments originating in India, Kenya, and Nigeria. The dataset is intended to support the study of gender-related discourse and stance, distinguishing gender-related comments from unrelated ones rather than focusing solely on harmful speech.

Because this is a single preprint abstract, it does not provide details on dataset size, annotation methodology, or evaluation results. The contribution, as described, is a more inclusive resource for studying how gender is discussed online in under-represented linguistic and geographic contexts.