Prompt-Based Approach for Czech Sentiment Analysis

2025-08-12

AI Summary

This paper addresses the poor zero-shot and few-shot performance of Czech in aspect-based sentiment analysis (ABSA) and sentiment classification. We propose the first prompt-based learning framework tailored for Czech. Methodologically, we adopt a unified sequence-to-sequence architecture for both tasks and pioneer the integration of target-domain pretraining with learnable soft prompts. Our contributions are twofold: (1) the construction of the first Czech-specific sentiment analysis prompt template suite; and (2) the co-optimization of domain-adaptive pretraining and prompt tuning. Experiments demonstrate that our approach significantly outperforms standard fine-tuning under few-shot settings and achieves a 28.6% F1-score improvement in zero-shot sentiment classification. This work establishes a transferable paradigm for sentiment analysis in low-resource languages.

Technology Category

Natural Language Processing: Prompt Engineering / PromptingMachine Learning: Transfer, Domain Adaptation, Multi-Task LearningKnowledge Representation and Reasoning: Common-Sense Reasoning

Application Category

Web Mining and Content Analysis: Large pretrained models with web dataSearch and Retrieval-Augmented AI: Multilingual and cross-lingual Web searchSemantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

This paper introduces the first prompt-based methods for aspect-based sentiment analysis and sentiment classification in Czech. We employ the sequence-to-sequence models to solve the aspect-based tasks simultaneously and demonstrate the superiority of our prompt-based approach over traditional fine-tuning. In addition, we conduct zero-shot and few-shot learning experiments for sentiment classification and show that prompting yields significantly better results with limited training examples compared to traditional fine-tuning. We also demonstrate that pre-training on data from the target domain can lead to significant improvements in a zero-shot scenario.

Problem

• First prompt-based Czech sentiment analysis methods
• Superiority of prompting over traditional fine-tuning
• Zero-shot and few-shot learning improvements

Innovation

• Prompt-based methods for Czech sentiment analysis
• Sequence-to-sequence models for aspect-based tasks
• Zero-shot and few-shot learning with prompting