|
--- |
|
model_creators: |
|
- Jordan Painter, Diptesh Kanojia |
|
tags: |
|
- generated_from_trainer |
|
metrics: |
|
- accuracy |
|
- precision |
|
- recall |
|
- f1 |
|
widget: |
|
- text: "I'm ecstatic my flight was just delayed" |
|
model-index: |
|
- name: bertweet-base-finetuned-SARC-DS |
|
results: [] |
|
--- |
|
# Utilising Weak Supervision to Create S3D: A Sarcasm Annotated Dataset |
|
This is the repository for the S3D dataset published at EMNLP 2022. The dataset can help build sarcasm detection models. |
|
|
|
# bertweet-base-finetuned-SARC-DS |
|
|
|
This model is a fine-tuned version of [vinai/bertweet-base](https://huggingface.co/vinai/bertweet-base) on the [SARC](https://metatext.io/datasets/self-annotated-reddit-corpus-(sarc)) dataset. |
|
It achieves the following results on the evaluation set: |
|
- Loss: 1.7094 |
|
- Accuracy: 0.7636 |
|
- Precision: 0.7637 |
|
- Recall: 0.7636 |
|
- F1: 0.7636 |
|
|
|
## Model description |
|
|
|
The given description for BERTweet by VinAI is as follows: <br> |
|
BERTweet is the first public large-scale language model pre-trained for English Tweets. BERTweet is trained based on the RoBERTa pre-training procedure. The corpus used to pre-train BERTweet consists of 850M English Tweets (16B word tokens ~ 80GB), containing 845M Tweets streamed from 01/2012 to 08/2019 and 5M Tweets related to the COVID-19 pandemic. |
|
<br> |
|
|
|
## Training and evaluation data |
|
|
|
This [vinai/bertweet-base](https://huggingface.co/vinai/bertweet-base) model was finetuned on the [SARC](https://metatext.io/datasets/self-annotated-reddit-corpus-(sarc)) dataset. The dataset is intended to help build sarcasm detection models. |
|
|
|
## Training procedure |
|
|
|
### Training hyperparameters |
|
|
|
The following hyperparameters were used during training: |
|
- learning_rate: 1e-05 |
|
- train_batch_size: 16 |
|
- eval_batch_size: 32 |
|
- seed: 43 |
|
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
|
- lr_scheduler_type: linear |
|
- num_epochs: 20 |
|
|
|
### Training results |
|
|
|
| Training Loss | Epoch | Step | Validation Loss | Accuracy | Precision | Recall | F1 | |
|
|:-------------:|:-----:|:------:|:---------------:|:--------:|:---------:|:------:|:------:| |
|
| 0.4978 | 1.0 | 44221 | 0.4899 | 0.7777 | 0.7787 | 0.7778 | 0.7775 | |
|
| 0.4413 | 2.0 | 88442 | 0.4833 | 0.7798 | 0.7803 | 0.7798 | 0.7797 | |
|
| 0.3943 | 3.0 | 132663 | 0.5387 | 0.7784 | 0.7784 | 0.7784 | 0.7784 | |
|
| 0.3461 | 4.01 | 176884 | 0.6184 | 0.7690 | 0.7701 | 0.7690 | 0.7688 | |
|
| 0.3024 | 5.01 | 221105 | 0.6899 | 0.7684 | 0.7691 | 0.7684 | 0.7682 | |
|
| 0.2653 | 6.01 | 265326 | 0.7805 | 0.7654 | 0.7660 | 0.7654 | 0.7653 | |
|
| 0.2368 | 7.01 | 309547 | 0.9066 | 0.7643 | 0.7648 | 0.7643 | 0.7642 | |
|
| 0.2166 | 8.01 | 353768 | 1.0548 | 0.7612 | 0.7620 | 0.7611 | 0.7610 | |
|
| 0.2005 | 9.01 | 397989 | 1.0649 | 0.7639 | 0.7639 | 0.7639 | 0.7639 | |
|
| 0.1837 | 10.02 | 442210 | 1.1805 | 0.7621 | 0.7624 | 0.7621 | 0.7621 | |
|
| 0.1667 | 11.02 | 486431 | 1.3017 | 0.7658 | 0.7659 | 0.7659 | 0.7658 | |
|
| 0.1531 | 12.02 | 530652 | 1.2947 | 0.7627 | 0.7628 | 0.7627 | 0.7627 | |
|
| 0.1377 | 13.02 | 574873 | 1.3877 | 0.7639 | 0.7639 | 0.7639 | 0.7639 | |
|
| 0.1249 | 14.02 | 619094 | 1.4468 | 0.7613 | 0.7616 | 0.7613 | 0.7612 | |
|
| 0.1129 | 15.02 | 663315 | 1.4951 | 0.7620 | 0.7621 | 0.7620 | 0.7620 | |
|
| 0.103 | 16.02 | 707536 | 1.5599 | 0.7619 | 0.7624 | 0.7619 | 0.7618 | |
|
| 0.0937 | 17.03 | 751757 | 1.6270 | 0.7615 | 0.7616 | 0.7615 | 0.7615 | |
|
| 0.0864 | 18.03 | 795978 | 1.6918 | 0.7622 | 0.7624 | 0.7622 | 0.7621 | |
|
| 0.0796 | 19.03 | 840199 | 1.7094 | 0.7636 | 0.7637 | 0.7636 | 0.7636 | |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.20.1 |
|
- Pytorch 1.10.1+cu111 |
|
- Datasets 2.3.2 |
|
- Tokenizers 0.12.1 |
|
|
