We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.CL

Change to browse by:

cs

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computation and Language

Title: Meta AI at Arabic Hate Speech 2022: MultiTask Learning with Self-Correction for Hate Speech Classification

Abstract: In this paper, we tackle the Arabic Fine-Grained Hate Speech Detection shared task and demonstrate significant improvements over reported baselines for its three subtasks. The tasks are to predict if a tweet contains (1) Offensive language; and whether it is considered (2) Hate Speech or not and if so, then predict the (3) Fine-Grained Hate Speech label from one of six categories. Our final solution is an ensemble of models that employs multitask learning and a self-consistency correction method yielding 82.7% on the hate speech subtask -- reflecting a 3.4% relative improvement compared to previous work.
Comments: Accepted at the 5th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT5/LREC 2022)
Subjects: Computation and Language (cs.CL)
Cite as: arXiv:2205.07960 [cs.CL]
  (or arXiv:2205.07960v1 [cs.CL] for this version)

Submission history

From: Badr AlKhamissi [view email]
[v1] Mon, 16 May 2022 19:53:16 GMT (68kb,D)

Link back to: arXiv, form interface, contact.