With the rapid growth of location-based services and the demand for intelligent personalization, location-based recommender systems (LBRS) have emerged as essential tools for improving user experience in physical environments. This study presents a comparative analysis of Transformer and LSTM architectures for predicting users' next locations based on historical POI sequences, contextual features, and temporal patterns. To address the challenge of contextual granularity, two representations - Fine-Grained and Coarse-Grained - are systematically compared. The Coarse-Grained approach utilizes BERT for semantic clustering of POI categories. Experimental results on real-world check-in data demonstrate that the LSTM model achieves superior performance with 13.6% higher NDCG@5 (0.2159 vs. 0.1901) and 15.7% better Precision@5 (0.0309 vs. 0.0267) compared to Transformer, while maintaining competitive accuracy through intelligent feature space reduction. According to the findings, in nearly all evaluations, leveraging contextual information in a coarse-grained manner consistently yielded better results than using it in a fine-grained manner. These findings offer practical solutions for scalable and accurate POI recommendation in tourism and smart mobility applications.