MetaGeo: A General Framework for Social User Geolocation Identification With Few-Shot Learning

Area of Science:

• Computer Science
• Social Computing
• Machine Learning

Background:

• User geolocation identification is crucial for diverse applications like disease surveillance and fake news detection.
• Current methods struggle with limited data and poor generalization to new geographic regions.

Purpose of the Study:

• To introduce MetaGeo, a meta-learning framework designed to enhance user geolocation prediction.
• To address challenges of data scarcity and regional generalization in geolocation tasks.

Main Methods:

• MetaGeo employs a meta-learning approach, learning prior distributions for rapid adaptation to new locations.
• It ensembles numerous mini-tasks and utilizes probabilistic inference to handle location uncertainty and task ambiguity.
• The framework is evaluated on three real-world datasets against state-of-the-art models.

Main Results:

• MetaGeo demonstrates superior performance in user geolocation prediction compared to benchmark models.
• The framework shows effectiveness in both known and unseen regions during training.
• Probabilistic inference successfully mitigates issues related to limited samples.

Conclusions:

• MetaGeo offers a robust and generalizable solution for social media user geolocation.
• The meta-learning and probabilistic inference approach significantly improves prediction accuracy and adaptability.
• This framework advances the field of geo-spatial social media analysis.