One Model, Many Cities: A Transferable Social Relationship Inference Framework for Human Mobility Data.

Area of Science:

• Computational Social Science
• Data Science
• Machine Learning

Background:

• Inferring social relationships from mobility data is vital for numerous applications.
• A significant challenge is the scarcity of large-scale trajectory datasets with ground-truth social ties, hindering deep model training.
• Existing methods struggle with generalization across diverse datasets.

Purpose of the Study:

• To develop a transferable framework for social relationship inference from mobility data.
• To enable models trained on one dataset to generalize to new, unseen datasets, even from different geographical locations.
• To overcome the limitations of data scarcity and improve the robustness of social relationship inference.

Main Methods:

• Proposed a transferable social relationship inference framework comprising two modules: Universal Social Relationship Classifier (USRC) and Spatial Embedding Transfer (SET).
• USRC is trained to infer social relationships from trajectory data based on meeting frequency and location popularity.
• SET aligns location embeddings to adapt pre-trained models to new datasets without requiring extra supervision.

Main Results:

• The proposed method achieved state-of-the-art performance in zero-shot social relationship inference across five public datasets.
• Outperformed unsupervised and some supervised approaches in inferring social ties from mobility data.
• The SET module significantly enhanced location embedding alignment compared to existing baseline methods.

Conclusions:

• The developed framework offers a robust and generalizable solution for social relationship inference from mobility data.
• The approach effectively addresses the challenge of limited labeled data by enabling cross-dataset generalization.
• This work advances the field by providing a method that performs well even without direct supervision on target datasets.