Modern recommender systems increasingly face the challenge of capturing dynamic and context-rich user preferences. Traditional collaborative filtering and content-based approaches often struggle to model temporal patterns and evolving user intentions. While Large Language Models (LLMs) have recently gained attention for their strong semantic understanding and reasoning capabilities, they are not inherently suited for modeling chronologically evolving preferences. Conversely, sequential models like Long Short-Term Memory (LSTM) networks excel at capturing temporal user dynamics but lack the semantic depth needed for comprehensive recommendations.

In this study, I propose DUALRec (Dynamic User-Aware Language-based Recommender), a hybrid recommendation framework that combines the strengths of both approaches. The LSTM component models users’ evolving preferences based on their viewing history, while fine-tuned LLM variants (DeepSeek and Mistral) incorporate these insights to generate context-aware movie recommendations. Experimental results on the ML-1M dataset demonstrate that DUALRec significantly outperforms a range of baseline models, evaluated using Hit Rate (HR@k), Normalized Discounted Cumulative Gain (NDCG@k), and genre similarity metrics. This research contributes to advancing next-generation recommendation systems by bridging temporal modeling and semantic reasoning.