Social Network Location Inference via Geometric Median
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for inferring user location in online social media platforms are limited by requiring geographic references in user-generated text, relying on known locations for social network inference, and having low coverage and high error rates, especially when users lack self-reported location information.
Innovation Solution
A system that extracts social networks from multiple platforms, uses geometric median metrics to estimate user locations based on social relationships, and merges location data from different platforms to maximize inference accuracy, incorporating a social triangle heuristic to filter relevant relationships and reduce noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If text content analysis is used for location inference, then location accuracy is improved for users with geographic references, but coverage is limited to only those users who generated text containing geographic references
Solution Approach 1:
The patent uses social network relationships as an intermediary to transfer location information from users with known locations to users without geographic references. Instead of directly analyzing text content for each user, the system mediates through the social graph, allowing location inference to propagate through friendship connections, thereby expanding coverage to users who lack geographic content in their posts.
Solution Approach 2:
The patent transitions from a single-dimension approach (text content analysis) to a multi-dimensional approach by incorporating social network relationships as an additional dimension. This allows the system to infer locations not only from geographic references in text but also from the spatial distribution of a user's social connections, thereby improving both coverage and accuracy.
2Quantity of substance
If social network inference is used to estimate user location, then coverage is improved, but the approach requires both users' locations be known to estimate social relationship
Solution Approach 1:
The patent performs preliminary location inference using available geographic references and text content before attempting social network-based inference. This preliminary action creates an initial set of known locations that can then serve as seeds for propagating location information through the social network, breaking the circular dependency where known locations are required to estimate social relationships.
Solution Approach 2:
The system implements an iterative feedback mechanism where initially inferred locations are used to improve social relationship estimates, which in turn enhance location inference accuracy. The process feeds results back into the system, allowing the model to progressively improve both location estimates and social relationship understanding without requiring complete prior knowledge.
3Ease of operation
If standard label propagation is used for location inference, then computational simplicity is maintained, but coverage is limited and accuracy is reduced
Solution Approach 1:
The patent segments the location inference process into multiple distinct stages: text content analysis, social network relationship extraction, iterative location estimation, and refinement. This segmentation allows each component to be optimized independently while maintaining overall computational efficiency, achieving both improved coverage and accuracy without sacrificing operational simplicity.
4Quantity of substance
If self provided location information is used, then coverage is improved to over 90% of users, but error rate becomes high when location data is sparse or inaccurate
Solution Approach 1:
The system uses feedback mechanisms to iteratively refine location estimates by comparing self-provided location information with locations inferred from social network relationships. Discrepancies between these sources trigger refinement processes that adjust the location estimates, thereby reducing error rates while maintaining the high coverage achieved through self-provided data.
Solution Approach 2:
The patent creates a composite location estimation approach that combines multiple data sources (self-provided location information, text content geographic references, and social network relationship-based inference) into a unified location estimate. This composite approach leverages the strengths of each source while compensating for their individual weaknesses, particularly reducing error rates in cases where self-provided data is sparse or inaccurate.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Described is a system for inferring the location of users in online social media platforms rising social network analysis, A social network is first extracted from data from at least one social media platform. A mapping is generated from each user to the user's estimated geographical location in the social network, resulting in an estimated location mapping. A mapping is generated from each user to the user's known geographical location, if known, resulting in a known location mapping. The estimated location mapping is updated to match the known location mapping. The location for each user j in the known location mapping is updated in a current estimated location mapping. The final geographical locations of users connected with / are estimated using a geometric median metric. Finally, the final estimated geographical locations of users connected with j are mapped into the social network.