This paper investigates the application of Neural Architecture Search (NAS) for optimizing sentiment analysis models, motivated by the need for more accurate and computationally efficient architectures in natural language processing tasks. Traditional NAS approaches have shown promising results in previous studies, but they are often time-intensive and computationally demanding, limiting their adoption. In this proposed method, we evaluate two NAS strategies, random search and evolutionary algorithm, which achieved respectable accuracy on the IMDB and Yelp datasets. Building on the insights from these individual strategies, we propose a hybrid NAS approach that leverages both random search and evolutionary principles to optimize model performance. According to the results, the hybrid approach achieved high accuracy with less processing time, consistently outperforming both independent approaches in terms of accuracy and computational economy. This study demonstrates the promise of hybrid NAS approaches for sentiment analysis, providing more scalable models with limited resources.