Tox-BART: Leveraging Toxicity Attributes for Explanation Generation of Implicit Hate Speech
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2024
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866916277708128256 |
|---|---|
| author | Yadav, Neemesh Masud, Sarah Goyal, Vikram Goyal, Vikram Akhtar, Md Shad Chakraborty, Tanmoy |
| author_facet | Yadav, Neemesh Masud, Sarah Goyal, Vikram Goyal, Vikram Akhtar, Md Shad Chakraborty, Tanmoy |
| contents | Employing language models to generate explanations for an incoming implicit hate post is an active area of research. The explanation is intended to make explicit the underlying stereotype and aid content moderators. The training often combines top-k relevant knowledge graph (KG) tuples to provide world knowledge and improve performance on standard metrics. Interestingly, our study presents conflicting evidence for the role of the quality of KG tuples in generating implicit explanations. Consequently, simpler models incorporating external toxicity signals outperform KG-infused models. Compared to the KG-based setup, we observe a comparable performance for SBIC (LatentHatred) datasets with a performance variation of +0.44 (+0.49), +1.83 (-1.56), and -4.59 (+0.77) in BLEU, ROUGE-L, and BERTScore. Further human evaluation and error analysis reveal that our proposed setup produces more precise explanations than zero-shot GPT-3.5, highlighting the intricate nature of the task. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2406_03953 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Tox-BART: Leveraging Toxicity Attributes for Explanation Generation of Implicit Hate Speech Yadav, Neemesh Masud, Sarah Goyal, Vikram Goyal, Vikram Akhtar, Md Shad Chakraborty, Tanmoy Computation and Language Employing language models to generate explanations for an incoming implicit hate post is an active area of research. The explanation is intended to make explicit the underlying stereotype and aid content moderators. The training often combines top-k relevant knowledge graph (KG) tuples to provide world knowledge and improve performance on standard metrics. Interestingly, our study presents conflicting evidence for the role of the quality of KG tuples in generating implicit explanations. Consequently, simpler models incorporating external toxicity signals outperform KG-infused models. Compared to the KG-based setup, we observe a comparable performance for SBIC (LatentHatred) datasets with a performance variation of +0.44 (+0.49), +1.83 (-1.56), and -4.59 (+0.77) in BLEU, ROUGE-L, and BERTScore. Further human evaluation and error analysis reveal that our proposed setup produces more precise explanations than zero-shot GPT-3.5, highlighting the intricate nature of the task. |
| title | Tox-BART: Leveraging Toxicity Attributes for Explanation Generation of Implicit Hate Speech |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2406.03953 |