Social Link Inference via Multiview Matching Network From Spatiotemporal Trajectories

Area of Science:

• Computer Science
• Social Network Analysis

Background:

• Social link inference is crucial for applications like friend recommendation in location-aware social networks (LSNs).
• Existing methods often overlook the temporal dynamics of user trajectories, limiting prediction accuracy.
• Effective fusion of social, spatial, and temporal factors for user matching remains a challenge.

Purpose of the Study:

• To propose a novel multiview matching network (MVMN) for social link inference in LSNs.
• To effectively model and integrate social, spatial, and temporal user data for improved link prediction.
• To address the limitations of prior studies by incorporating the temporal factor in user trajectory analysis.

Main Methods:

• Developed a Multiview Matching Network (MVMN) that treats social, spatial, and temporal data as distinct views.
• Implemented specialized matching modules for location, time-series, and social relations.
• Fused view-specific representations for comprehensive user matching and link inference.

Main Results:

• The MVMN demonstrated superior performance in link prediction and sequence matching compared to baseline methods.
• Experiments on two real-world datasets validated the effectiveness of the proposed approach.
• The contribution of individual components, including the temporal factor, was confirmed.

Conclusions:

• The proposed MVMN effectively integrates spatial and temporal information for enhanced social link inference in LSNs.
• Incorporating temporal dynamics significantly improves the accuracy of predicting unobserved user links.
• This work provides a robust framework for user matching and link prediction by leveraging multiple data views.