FCKT: Fine-Grained Cross-Task Knowledge Transfer with Semantic Contrastive Learning for Targeted Sentiment Analysis

TL;DR

The paper tackles targeted sentiment analysis (TSA) by proposing FCKT, a fine-grained cross-task knowledge transfer framework that explicitly couples aspect extraction and sentiment prediction. It introduces token-level semantic contrastive learning to align start and end tokens of the same aspect, mitigating error propagation, and an alternating supervision strategy that blends real labels with transferred predictions to strengthen sentiment classification. The approach is backed by end-to-end training with a boundary-based extraction mechanism and a boundary-aggregated sentiment predictor, yielding improvements over strong baselines and LLMs on three real-world datasets. The results highlight the practical value of fine-grained transfer and token-level supervision for TSA, suggesting a viable path for robust task-specific methods in the era of large language models, with clear guidance on hyperparameters and model design.

Abstract

In this paper, we address the task of targeted sentiment analysis (TSA), which involves two sub-tasks, i.e., identifying specific aspects from reviews and determining their corresponding sentiments. Aspect extraction forms the foundation for sentiment prediction, highlighting the critical dependency between these two tasks for effective cross-task knowledge transfer. While most existing studies adopt a multi-task learning paradigm to align task-specific features in the latent space, they predominantly rely on coarse-grained knowledge transfer. Such approaches lack fine-grained control over aspect-sentiment relationships, often assuming uniform sentiment polarity within related aspects. This oversimplification neglects contextual cues that differentiate sentiments, leading to negative transfer. To overcome these limitations, we propose FCKT, a fine-grained cross-task knowledge transfer framework tailored for TSA. By explicitly incorporating aspect-level information into sentiment prediction, FCKT achieves fine-grained knowledge transfer, effectively mitigating negative transfer and enhancing task performance. Experiments on three datasets, including comparisons with various baselines and large language models (LLMs), demonstrate the effectiveness of FCKT. The source code is available on https://github.com/cwei01/FCKT.

Figures (6)

• Figure 1: Illustration comparing coarse-grained and fine-grained transfer methods. Coarse-grained transfer assumes uniform sentiment polarity for related aspects, while fine-grained transfer captures nuanced aspect-sentiment relationships for precise alignment.
• Figure 2: A depiction of the peopose FCKT framework.
• Figure 3: An example of the sentence splitting process.
• Figure 4: The results of FCKT w.r.t different parameter $\xi$ on two datasets. "FCKT w/o RAI" refers to removing real aspect inputs.
• Figure 5: The F1 score of FCKT w.r.t varying parameter $h$ and $\lambda$.