A Time-Location Theme Mining Algorithm Based on R-Tree and User Attention for Personalized Recommendation

This paper addresses the limitations of existing theme mining algorithms in extracting user-preferred themes from time-location data for personalized recommendation. We propose a Time-Location Theme Mining algorithm based on R-tree and user Attention (named as TLTMRA). TLTMRA combines mesh and R-Tree structures for efficient theme data processing and considers both the overall importance of themes and user attention to theme objects. Experimental results on real-world datasets demonstrate that TLTMRA outperforms state-of-the-art methods in terms of storage overhead, theme validity, and recommendation efficiency. The proposed algorithm achieves up to 59% theme validity and significantly reduces storage and computation costs compared to baseline methods. This work contributes to the development of effective and efficient personalized recommendation systems leveraging time-location data. Future research directions include extending the proposed method to other data types and recommendation scenarios and further optimizing the algorithm for large-scale applications.