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 on the 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:
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.

Training and evaluation data

This vinai/bertweet-base model was finetuned on the 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

surrey-nlp
/

bertweet-base-finetuned-SARC-DS