Guardado en:
| Autores principales: | , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2024
|
| Materias: | |
| Acceso en línea: | https://arxiv.org/abs/2410.05085 |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866909338749108224 |
|---|---|
| author | Bogaert, Jeremie de Marneffe, Marie-Catherine Descampe, Antonin Escouflaire, Louis Fairon, Cedrick Standaert, Francois-Xavier |
| author_facet | Bogaert, Jeremie de Marneffe, Marie-Catherine Descampe, Antonin Escouflaire, Louis Fairon, Cedrick Standaert, Francois-Xavier |
| contents | Large language models (LLMs) perform very well in several natural language processing tasks but raise explainability challenges. In this paper, we examine the effect of random elements in the training of LLMs on the explainability of their predictions. We do so on a task of opinionated journalistic text classification in French. Using a fine-tuned CamemBERT model and an explanation method based on relevance propagation, we find that training with different random seeds produces models with similar accuracy but variable explanations. We therefore claim that characterizing the explanations' statistical distribution is needed for the explainability of LLMs. We then explore a simpler model based on textual features which offers stable explanations but is less accurate. Hence, this simpler model corresponds to a different tradeoff between accuracy and explainability. We show that it can be improved by inserting features derived from CamemBERT's explanations. We finally discuss new research directions suggested by our results, in particular regarding the origin of the sensitivity observed in the training randomness. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2410_05085 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Explanation sensitivity to the randomness of large language models: the case of journalistic text classification Bogaert, Jeremie de Marneffe, Marie-Catherine Descampe, Antonin Escouflaire, Louis Fairon, Cedrick Standaert, Francois-Xavier Computation and Language Large language models (LLMs) perform very well in several natural language processing tasks but raise explainability challenges. In this paper, we examine the effect of random elements in the training of LLMs on the explainability of their predictions. We do so on a task of opinionated journalistic text classification in French. Using a fine-tuned CamemBERT model and an explanation method based on relevance propagation, we find that training with different random seeds produces models with similar accuracy but variable explanations. We therefore claim that characterizing the explanations' statistical distribution is needed for the explainability of LLMs. We then explore a simpler model based on textual features which offers stable explanations but is less accurate. Hence, this simpler model corresponds to a different tradeoff between accuracy and explainability. We show that it can be improved by inserting features derived from CamemBERT's explanations. We finally discuss new research directions suggested by our results, in particular regarding the origin of the sensitivity observed in the training randomness. |
| title | Explanation sensitivity to the randomness of large language models: the case of journalistic text classification |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2410.05085 |