|
--- |
|
language: |
|
- en |
|
tags: |
|
- intent detection |
|
license: "other" |
|
datasets: |
|
- ibm/vira-intents |
|
metrics: |
|
- accuracy |
|
widget: |
|
- text: "Should I be concerned about side effects of the vaccine if I'm breastfeeding?} & Is breastfeeding safe with the vaccine" |
|
example_title: "Breastfeeding" |
|
- text: "Does the vaccine prevent transmission?" |
|
example_title: "Transmission" |
|
- text: "Will the vaccine make me sterile or infertile? " |
|
example_title: "Infertility" |
|
--- |
|
|
|
## Model Description |
|
This model is based on RoBERTa large (Liu, 2019), fine-tuned on a dataset of intent expressions available [here](https://research.ibm.com/haifa/dept/vst/debating_data.shtml) and also on 🤗 Transformer datasets hub [here](https://huggingface.co/datasets/ibm/vira-intents). |
|
|
|
The model was created as part of the work described in [Benchmark Data and Evaluation Framework for Intent Discovery Around COVID-19 Vaccine Hesitancy |
|
](https://arxiv.org/abs/2205.11966). The model is released under the Community Data License Agreement - Sharing - Version 1.0 ([link](https://cdla.dev/sharing-1-0/)), If you use this model, please cite our paper. |
|
|
|
The official GitHub is [here](https://github.com/IBM/vira-intent-discovery). The script used for training the model is [trainer.py](https://github.com/IBM/vira-intent-discovery/blob/master/trainer.py). |
|
|
|
|
|
## Training parameters |
|
1. base_model = 'roberta-large' |
|
1. learning_rate=5e-6 |
|
1. per_device_train_batch_size=16, |
|
1. per_device_eval_batch_size=16, |
|
1. num_train_epochs=15, |
|
1. load_best_model_at_end=True, |
|
1. save_total_limit=1, |
|
1. save_strategy='epoch', |
|
1. evaluation_strategy='epoch', |
|
1. metric_for_best_model='accuracy', |
|
1. seed=123 |
|
|
|
## Data collator |
|
DataCollatorWithPadding |
|
|