README.md

---
license: mit
tags:
- NLP
datasets:
- Yaxin/SemEval2014Task4Raw
metrics:
- f1
- precision
- recall
pipeline_tag: text2text-generation
---

# atsc_tk-instruct-base-def-pos-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

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}
}
```