EECSA: Enhancing Textual Sentiment Analysis through Emotion-Enriched Contextual Integration and Self-Adaptive Prompting Optimization

Textual sentiment analysis remains a crucial task in natural language processing (NLP) that focuses on identifying sentiments towards specific entities within a t ext. Recently, large language models (LLMs) have demonstrated remarkable capabilities in understanding semantics and logical reasoning. However, LLMs often face challenges such as hallucinations, which can lead to inaccurate sentiment determination for target entities. Additionally, current methods frequently fail to fully exploit the extensive prior knowledge embedded within LLMs, resulting in suboptimal sentiment classification. To address these challenges, we propose the Emotion-Enriched Contextual Sentiment Analysis (EECSA) framework. EECSA integrates unique emotion-enriched constraints, backgrounds, and analogical reasoning to mitigate LLM hallucinations and employs self-adaptive bootstrap instructions optimization to further enhance LLM predictions. Specifically, constraints are constructed through a linguistic consistency mechanism and multiple inquiry rounds; background descriptions are incorporated as prompts for LLM inference; and analogies are generated by LLMs to enhance textual sentiment analysis. EECSA further refines initial LLM instructions through adaptive iterative optimization using a random search bootstrap algorithm, maximizing the effectiveness of LLM prompting. Extensive experiments of zero-shot and few-shot learning with GPT-3.5-turbo across six public datasets validate the superiority of our EECSA framework, demonstrating performance that in some cases surpasses GPT-4.0, the most advanced large language model in the world. Code is available at: https://github.com/albert-jin/EECSA.