← Search

Data · dataset · 2026

Dataset for an Implicit Hate Speech Detection Method Based on LLM Encoding with Inference-Time Activation Intervention

Listed in ScienceDB

This dataset is a processed version of three publicly available datasets: TOXICN, LATENT-HATRED, and ISHATE.

Description

TOXICN contains data collected from Zhihu and Baidu Tieba and covers topics such as sexism, racism, regional prejudice, and anti-LGBTQ content, while LATENT-HATRED and ISHATE are English implicit hate speech datasets featuring covert expressions such as group references, stereotypes, metaphors, and sarcasm.

Links

Where it is published

Catalogue records · 1

Topics

Inferred from text
Artificial intelligence 68% · Audio 65%
Provenance · 1 source records, 13 field assertions
SourceKeyLast seenRaw
ScienceDB10.57760/sciencedb.4155511 d agoJSON v1
FieldAssertionExtractorEvidence
access_levelsource · scidb cnconnector:scidb_cn@1.0.0
concepts[field].anzsrc:group:4602enrichment · scidb cntaxonomy-embedding@1.0.0title+keywords+description (68%)
concepts[field].local:field:computer-science-aimapping · scidb cnconnector:scidb_cn@1.0.0
concepts[field].local:field:earth-environmentalmapping · scidb cnconnector:scidb_cn@1.0.0
concepts[field].local:field:engineeringmapping · scidb cnconnector:scidb_cn@1.0.0
concepts[field].local:field:humanitiesmapping · scidb cnconnector:scidb_cn@1.0.0
concepts[field].local:field:life-sciencesmapping · scidb cnconnector:scidb_cn@1.0.0
concepts[field].local:field:social-sciencemapping · scidb cnconnector:scidb_cn@1.0.0
concepts[modality].local:modality:audioenrichment · scidb cnkeyword-concept-rules@1.0.0title+description (65%)
descriptionsource · scidb cnconnector:scidb_cn@1.0.0/metadata/dc/description
licensesource · scidb cnconnector:scidb_cn@1.0.0/metadata/dc/rights
publication_datesource · scidb cnconnector:scidb_cn@1.0.0
titlesource · scidb cnconnector:scidb_cn@1.0.0/metadata/dc/title