Instruction Tuning for Few-Shot Aspect-Based Sentiment Analysis

TL;DR

This paper introduces a unified, instruction-tuned T5 model framework for aspect-based sentiment analysis that significantly improves few-shot learning performance across multiple sub-tasks.

Contribution

It proposes a multi-task, instruction-based fine-tuning approach for ABSA, covering all sub-tasks and the full quadruple prediction, enhancing few-shot capabilities.

Findings

Achieves an 8.29 F1 score improvement in few-shot settings.

Outperforms existing methods on multiple benchmark datasets.

Demonstrates the effectiveness of instruction tuning for complex NLP tasks.

Abstract

Aspect-based Sentiment Analysis (ABSA) is a fine-grained sentiment analysis task which involves four elements from user-generated texts: aspect term, aspect category, opinion term, and sentiment polarity. Most computational approaches focus on some of the ABSA sub-tasks such as tuple (aspect term, sentiment polarity) or triplet (aspect term, opinion term, sentiment polarity) extraction using either pipeline or joint modeling approaches. Recently, generative approaches have been proposed to extract all four elements as (one or more) quadruplets from text as a single task. In this work, we take a step further and propose a unified framework for solving ABSA, and the associated sub-tasks to improve the performance in few-shot scenarios. To this end, we fine-tune a T5 model with instructional prompts in a multi-task learning fashion covering all the sub-tasks, as well as the entire quadruple prediction task. In experiments with multiple benchmark datasets, we show that the proposed multi-task prompting approach brings performance boost (by absolute 8.29 F1) in the few-shot learning setting.