LLaMA-Based Models for Aspect-Based Sentiment Analysis

TL;DR

This paper explores fine-tuning LLaMA-based large language models for aspect-based sentiment analysis, demonstrating that fine-tuned Orca~2 outperforms existing models across multiple datasets, while highlighting challenges in zero-shot scenarios.

Contribution

It is the first comprehensive evaluation of open-source LLaMA-based models for ABSA, showing fine-tuning significantly improves performance over zero-shot methods.

Findings

Fine-tuned Orca~2 surpasses state-of-the-art in all tasks.

Models underperform in zero-shot and few-shot settings.

Error analysis reveals key challenges in model performance.

Abstract

While large language models (LLMs) show promise for various tasks, their performance in compound aspect-based sentiment analysis (ABSA) tasks lags behind fine-tuned models. However, the potential of LLMs fine-tuned for ABSA remains unexplored. This paper examines the capabilities of open-source LLMs fine-tuned for ABSA, focusing on LLaMA-based models. We evaluate the performance across four tasks and eight English datasets, finding that the fine-tuned Orca~2 model surpasses state-of-the-art results in all tasks. However, all models struggle in zero-shot and few-shot scenarios compared to fully fine-tuned ones. Additionally, we conduct error analysis to identify challenges faced by fine-tuned models.