|
--- |
|
license: mit |
|
tags: |
|
- NLP |
|
datasets: |
|
- Yaxin/SemEval2014Task4Raw |
|
metrics: |
|
- f1 |
|
- precision |
|
- recall |
|
pipeline_tag: text2text-generation |
|
--- |
|
|
|
# atsc_tk-instruct-base-def-pos-neg-neut-combined |
|
This model is finetuned for the Aspect Term Sentiment Classification (ATSC) subtask. The finetuning was carried out by adding prompts of the form: |
|
- definition + 2 positive examples + 2 negative examples + 2 neutral examples |
|
|
|
The prompt is prepended onto each input review. It is important to note that **this model output was finetuned on samples from both laptops and restaurants domains.** |
|
The code for the official implementation of the paper [**InstructABSA: Instruction Learning for Aspect Based Sentiment Analysis**](https://arxiv.org/abs/2302.08624) can be |
|
found [here](https://github.com/kevinscaria/InstructABSA). |
|
|
|
For the ATSC subtask, this model has a competitive performance with the current SOTA. |
|
|
|
## Training data |
|
|
|
InstructABSA models are trained on the benchmark dataset for Aspect Based Sentiment Analysis tasks viz. SemEval 2014. This [dataset](https://alt.qcri.org/semeval2014/task4/index.php?id=data-and-tools) consists of reviews |
|
from laptops and restaurant domains and their corresponding aspect term and polarity labels. |
|
|
|
### BibTeX entry and citation info |
|
|
|
If you use this model in your work, please cite the following paper: |
|
|
|
```bibtex |
|
@inproceedings{Scaria2023InstructABSAIL, |
|
title={InstructABSA: Instruction Learning for Aspect Based Sentiment Analysis}, |
|
author={Kevin Scaria and Himanshu Gupta and Saurabh Arjun Sawant and Swaroop Mishra and Chitta Baral}, |
|
year={2023} |
|
} |
|
``` |
