Doing More with Less: Overcoming Data Scarcity for POI Recommendation via Cross-Region Transfer

Variability in social app usage across regions results a high skew of the quantity and quality check-in data collected, which turn is challenge for effective location recommender systems. In this paper, we present Axolotl (Automated cross Location-network Transfer Learning), novel method aimed at transferring preference models learned data-rich region to significantly boost recommendations data-scarce region. predominantly deploys two channels information transfer, (1) meta-learning based procedure using recommendation as well predictions, (2) lightweight unsupervised cluster-based transfer users locations with similar preferences. Both these work together synergistically achieve improved accuracy without any prerequisite overlapping minimal fine-tuning. We build on top twin graph-attention neural network model used capturing user- location-conditioned influences user-mobility graph each conduct extensive experiments 12 user mobility datasets U.S., Japan, Germany, 3 source 9 them (that have much sparsely recorded data) target regions. Empirically, show that achieves up 18% better performance than existing state-of-the-art methods all metrics.