Understanding the Influence of Data Characteristics on the Performance of Point-of-Interest Recommendation Algorithms

2023-11-13

Information Technology & Tourism

AI Summary

This study investigates the impact mechanisms of data characteristics—such as sparsity, geographical bias, and uneven spatiotemporal distribution—on the performance of point-of-interest (POI) recommendation in tourism scenarios. To address declining accuracy, novelty, and fairness, we propose the first quantitative attribution model linking data characteristics to recommendation performance, along with an interpretable evaluation framework and a POI data health diagnostic metric. Methodologically, we integrate multidimensional statistical analysis, causal inference modeling, and cross-algorithm benchmarking (including BERT4Rec, STGN, and LBSN2Vec) on Foursquare and Twitter datasets. Empirical results identify geographical bias and session-length imbalance as primary drivers of performance degradation. Our diagnostic framework improves algorithm selection accuracy by 37%, enabling data-aware, principled model deployment. The work establishes a foundation for diagnosing and mitigating data-induced biases in location-based recommendation systems.

Problem

• POI Recommendation
• Data Characteristics
• Tourism Industry

Innovation

• POI Recommendation
• Evaluation Framework
• Data Characteristics Impact