Leveraging Multi-aspect Time-related Influence in Location Recommendation

Saeid Hosseini, Hongzhi Yin, Xiaofang Zhou, Shazia W. Sadiq

Abstract

Point-Of-Interest (POI) recommendation aims to mine a user's visiting history and find her/his potentially preferred places. Although location recommendation methods have been studied and improved pervasively, the challenges w.r.t employing various influences including temporal aspect still remain. Inspired by the fact that time includes numerous granular slots (e.g. minute, hour, day, week and etc.), in this paper, we define a new problem to perform recommendation through exploiting all diversified temporal factors. In particular, we argue that most existing methods only focus on a limited number of time-related features and neglect others. Furthermore, considering a specific granularity (e.g. time of a day) in recommendation cannot always apply to each user or each dataset. To address the challenges, we propose a probabilistic generative model, named after Multi-aspect Time-related Influence (MATI) to promote POI recommendation. We also develop a novel optimization algorithm based on Expectation Maximization (EM). Our MATI model firstly detects a user's temporal multivariate orientation using her check-in log in Location-based Social Networks(LBSNs). It then performs recommendation using temporal correlations between the user and proposed locations. Our method is adaptable to various types of recommendation systems and can work efficiently in multiple time-scales. Extensive experimental results on two large-scale LBSN datasets verify the effectiveness of our method over other competitors.

Introduction

Location-based Social Networks (LBSNs) like Foursquare and Yelp provide rich data for understanding human mobility. However, recommending the next Point-of-Interest (POI) is challenging due to extreme data sparsity; users typically visit only a fraction of available venues. While geographical and social factors are commonly used to mitigate this, temporal influence is often oversimplified. Most existing models focus on a single granularity, such as "hour of the day," ignoring the multi-layered nature of human behavior where a venue's popularity might vary by both hour and day of the week.

Methodology: The MATI Model

The proposed Multi-aspect Time-related Influence (MATI) model is a probabilistic generative framework designed to capture overlapping temporal influences. It consists of three primary components:

  1. Temporal Subset Property (TSP): Recognizes the hierarchical nature of time (minutes $\subset$ hours $\subset$ days $\subset$ weeks) to model multivariate correlations.
  2. Multi-aspect Temporal Slabs: To address sparsity, the model uses Hierarchical Agglomerative Clustering (HAC) to group similar time slots into "slabs" (e.g., late-night hours). This smoothing allows for more robust predictions even when specific hourly data is thin.
  3. Hybrid Decision Method: A "gatekeeper" mechanism assesses a user's temporal sensitivity. If a user's behavior is erratic or they are in a "cold start" phase, the system falls back to standard collaborative filtering.
Conceptual diagram of User Act Margin showing weekday ($w_d$) and weekend ($w_e$) activity fluctuations.

The model parameters are optimized using the Expectation-Maximization (EM) algorithm to infer latent temporal factors from check-in logs.

Key Insights and Empirical Analysis

A core observation of this work is the "Weekday/Weekend" divide. The authors found that a significant majority of users and venues exhibit a strong orientation toward specific parts of the week.

Distribution of weekday/weekend deviation for users in the Foursquare dataset.

As shown in the empirical data for Foursquare, many users have a high "Absolute Rate of Deviation," indicating consistent routines that favor either weekdays or weekends. Furthermore, temporal patterns are continuous; behaviors at adjacent hours are more similar than those further apart. MATI's "Temporal Slabs" capture this continuity effectively, allowing the model to leverage a location's "popular time" as a strong predictor of a visit.

Experimental Evaluation

MATI was evaluated using two large-scale datasets:

  • Brightkite: ~4.5 million check-ins, characterized by high sparsity (53% cold-start users).
  • Foursquare: ~483,000 check-ins, offering a denser user-visit profile.

MATI consistently outperformed baselines like User-Based Collaborative Filtering (UBCF) and models incorporating geographical/social factors (USG). In "Recommendation@5" tasks, MATI improved success rates by 4.1% on Brightkite and 6.6% on Foursquare compared to the best-performing baseline.

Conclusion and Future Work

MATI demonstrates that integrating multiple temporal scales significantly enhances POI recommendation accuracy and robustness against data sparsity. The hybrid approach ensures that temporal predictions are only made when a user's routine is sufficiently stable. Future work will focus on dynamic modeling to account for routine shifts (e.g., holidays) and deeper contextual smoothing between overlapping temporal factors.

Summary updated

Abstract

Point-Of-Interest (POI) recommendation aims to mine a user's visiting history and find her/his potentially preferred places. Although location recommendation methods have been studied and improved pervasively, the challenges w.r.t employing various influences including temporal aspect still remain. Inspired by the fact that time includes numerous granular slots (e.g. minute, hour, day, week and etc.), in this paper, we define a new problem to perform recommendation through exploiting all diversified temporal factors. In particular, we argue that most existing methods only focus on a limited number of time-related features and neglect others. Furthermore, considering a specific granularity (e.g. time of a day) in recommendation cannot always apply to each user or each dataset. To address the challenges, we propose a probabilistic generative model, named after Multi-aspect Time-related Influence (MATI) to promote POI recommendation. We also develop a novel optimization algorithm based on Expectation Maximization (EM). Our MATI model firstly detects a user's temporal multivariate orientation using her check-in log in Location-based Social Networks(LBSNs). It then performs recommendation using temporal correlations between the user and proposed locations. Our method is adaptable to various types of recommendation systems and can work efficiently in multiple time-scales. Extensive experimental results on two large-scale LBSN datasets verify the effectiveness of our method over other competitors.
BibTeX
@article{hosseini2019leveraging,
  author = {Saeid Hosseini and Hongzhi Yin and Xiaofang Zhou and Shazia W. Sadiq},
  title = {{Leveraging Multi-aspect Time-related Influence in Location Recommendation}},
  journal = {World Wide Web},
  year = {2019},
  doi = {10.1007/s11280-018-0573-2},
}
Figure