Project · NLP · Sequence labeling

Hate span detection with BERT & CRF

Combined BERT representations with a conditional random field to identify hateful spans within sentences.

Span-level prediction

The task identifies which portions of a sentence contain hateful content. This requires more granular predictions than assigning a single label to the entire sentence.

Modeling approach

Used BERT and conditional random fields for sequence labeling, combining contextual representations with a model of dependencies between output labels.

BERT
Conditional random fields
Sequence labeling