---
license: apache-2.0
metrics:
- accuracy
- f1
pipeline_tag: text-classification
tags:
- irony
language:
- it
---

This model is a fine-tuned version of [AlBERTo](m-polignano-uniba/bert_uncased_L-12_H-768_A-12_italian_alb3rt0) Italian model on the following datasets annotated with the presence of irony:

- [IronITA 2018](https://live.european-language-grid.eu/catalogue/corpus/7372)
- [Sarcastic Hate Speech dataset](https://github.com/simonasnow/Sarcastic-Hate-Speech)
- SENTIPOLC [2014](https://live.european-language-grid.eu/catalogue/corpus/7480)/[2016](https://live.european-language-grid.eu/catalogue/corpus/7479)

It was tested on IronITA test set obtaining the following results:

- macro F1: 0.79
- accuracy: 0.79
- precision of positive class: 0.77
- recall of positive class: 0.84
- F1 of positive class: 0.80

Training hyperparameters:

- learning_rate: 2e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- optimizer: Adam

Framework versions:

- Transformers 4.30.2
- Pytorch 2.1.2
- Datasets 2.19.0
- Accelerate=0.30.0-