Geo-Semantic Hybrid Model For Recommending Tourist Accommodations Based On Real Data and Empirical Evaluation

The growth of digital tourist accommodation platforms has increased the com-plexity of decision-making, creating a need for recommendation systems that in-tegrate multiple relevant dimensions. This paper proposes a hybrid geo-semantic model for recommending tourist accommodation that explicitly combines concep-tual similarity and geographical proximity within a unified scoring function. The model was developed and evaluated on a real dataset consisting of 13,204 ac-commodations and 467 descriptive variables. Additionally, the impact of dimen-sionality reduction was analyzed using Principal Component Analysis (PCA), obtaining 297 representative components without significantly affecting the se-mantic coherence of the system. The results show that the model maintains aver-age similarity levels comparable to those observed in a widely used commercial platform (0.774 vs. 0.758), while substantially reducing the average geographical distance of the recommendations (0.33 km vs. 1.94 km). This difference trans-lates into a higher average hybrid score, consistent with the defined objective function. The findings support the methodological viability of the proposed hy-brid approach and show that the explicit integration of semantic and spatial di-mensions constitutes a structured alternative for tourism recommendation systems based on real data.