<p>Generative artificial intelligence (GenAI) has rapidly advanced in recent years, primarily focusing on problem-solving tasks. However, its potential for sentiment analysis and applications in psychological contexts remains underexplored. This paper presents a comprehensive review of recent developments in generative AI for sentiment analysis, covering transformer-based models, variational autoencoders, multimodal approaches, generative adversarial networks, and large language models such as GPT, BERT, Claude, and Falcon. We analyze their strengths and limitations, addressing challenges including data dependency, imbalanced dataset, scalability, robustness, and interpretability. At the same time, future research directions are highlighted, such as low-resource and multilingual sentiment analysis, multimodal learning, reinforcement learning with human feedback, and domain-specific modeling. By integrating current findings, this study aims to guide future research toward building sentiment analysis systems that are not only technically robust but also beneficial in psychological and social domains.</p>