Sentiment analysis is a critical Natural Language Processing (NLP) task that categorizes text into positive, negative or neutral sentiments. While sentiment analysis has been extensively studied for English, research on Romanian sentiment analysis remains limited, primarily due to the scarcity of linguistic resources and specialized tools. To address this gap, AugRoSent (Augmentation for Romanian Sentiment Analysis) introduces a dataset augmentation framework designed to enhance the robustness and accuracy of Romanian sentiment analysis models. Three distinct augmentation techniques are applied within AugRoSent: Easy Data Augmentation (EDA), leveraging simple textual operations; text generation using a Romanian-specific open-source Large Language Model (OpenLLM-Ro); and fine-tuned GPT-2 models adapted to Romanian linguistic nuances. Each method’s effectiveness is evaluated on a benchmark Romanian sentiment analysis dataset using the standard metric accuracy, alongside a series of qualitative metrics. Experimental results demonstrate that dataset augmentation improves model performance. These findings highlight the value of advanced language modeling approaches for resource-constrained languages and provide insights into selecting optimal augmentation strategies for Romanian NLP applications.