Beyond Check-in Counts: Redefining Popularity for POI Recommendation with Users and Recency

2024-07-07

arXiv.org

AI Summary

Existing POI recommendation methods predominantly define popularity based on static check-in counts, neglecting temporal decay, frequency variation, and user diversity—leading to biased recommendations. To address this, we propose a dynamic POI popularity modeling framework that jointly incorporates three dimensions: check-in timestamp, frequency, and number of distinct users. Specifically, we introduce a time-decay function to capture behavioral recency, design a weighted aggregation mechanism to fuse heterogeneous signals, and explicitly model user coverage breadth to enhance recommendation diversity. The resulting dynamic popularity metric is model-agnostic and seamlessly integrates into mainstream sequential (e.g., BERT4Rec) and spatiotemporal graph-based (e.g., STGN) recommenders. Extensive experiments on multiple real-world trajectory datasets demonstrate consistent improvements: average gains of 12.3% in Recall@10 and NDCG@10 over count-based baselines, validating both the effectiveness and generalizability of our dynamic popularity modeling for next-POI recommendation.

Technology Category

• Data Mining & Knowledge Management: Recommender Systems
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• User Modeling, Personalization and Recommendation: Metrics for user behavior and evaluating success
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Web Mining and Content Analysis: Content-based information diffusion

Abstract

The next POI (point of interest) recommendation aims to predict users' immediate future movements based on their prior records and present circumstances, which will be very beneficial to service providers as well as users. The popularity of the POI over time is one of the primary deciding factors for choosing the next POI to visit. The majority of research in recent times has paid more attention to the number of check-ins to define the popularity of a point of interest, disregarding the temporal impact or number of people checking in for a particular POI. In this paper, we propose a recency-oriented definition of popularity that takes into account the temporal effect on POI's popularity, the number of check-ins, as well as the number of people who registered those check-ins. Thus, recent check-ins get prioritized with more weight compared to the older ones. Experimental results demonstrate that performance is better with recency-aware popularity definitions for POIs than with solely check-in count-based popularity definitions.

Problem

• Location-based Recommendation
• Temporal Influence
• Popularity Bias

Innovation

• Time-sensitive recommendation
• Check-in frequency
• Recent check-in emphasis