This paper presents a novel approach to recommendation systems by integrating large language models (LLMs) to address the semantic understanding limitations of traditional collaborative filtering and content-based methods. We propose LLMRec, a hybrid recommendation framework that leverages LLMs’ natural language understanding capabilities to capture nuanced user preferences and item attributes. Through extensive experiments on multiple datasets, we demonstrate that LLMRec outperforms state-of-the-art recommendation models across various evaluation metrics, with particularly significant improvements in cold-start scenarios and recommendation diversity. Our findings suggest that LLMs can effectively bridge the semantic gap in recommendation systems while maintaining computational efficiency through strategic integration approaches.