Social Media User Location Estimation via KL-Divergence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating user locations in social media, such as microblogs, face limitations due to the rarity of geographical names in posted text and imprecision in estimating locations based on regional words and user relationships.
Innovation Solution
A method that identifies local events attracting attention in a region and estimates the residence of users who mention these events by computing base and event distributions using KL-divergence, associating users with locations based on the degree of relationship and localization, and using a placename dictionary to determine latitude and longitude.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If location estimation is performed based on regionality of words or user relationships, then location estimation can be performed for users without filled-in location information, but the estimation precision is insufficient
Solution Approach 1:
The system pre-computes base distributions of words and user relationships across different geographic regions before actual location estimation is needed. These pre-computed distributions serve as reference data that enables rapid and accurate location estimation when a user's location is unknown, resolving the contradiction by preparing estimation criteria in advance rather than computing them on-demand during the estimation process
Solution Approach 2:
The patent introduces base distributions as an intermediary layer between raw social media data and location estimation results. These base distributions act as a mediator that translates user behavior patterns (words used, relationships formed) into probabilistic location information, thereby improving estimation accuracy without requiring direct geographic coordinates from users
2Measurement precision
If GPS geotag information is added to user posts, then precise location information can be obtained, but user privacy is compromised and not all users provide this information
Solution Approach 1:
The patent replaces the mechanical GPS-based location tracking system with an information-theoretic approach using KL-divergence to compare base distributions. Instead of directly collecting and processing geographic coordinates (mechanical system), the system uses probabilistic distribution comparisons to infer locations, thereby maintaining precision while reducing privacy intrusion
Solution Approach 2:
Base distributions serve as an intermediary that indirectly captures location information through analysis of user behavior patterns rather than directly collecting sensitive geographic data. This intermediary layer allows the system to obtain location estimates without users explicitly providing or exposing their precise location information
3Loss of information
If text analysis is performed to extract geographical names from user posts, then location information can be derived from posted content, but geographical names are rarely contained in user text
Solution Approach 1:
The system pre-computes base distributions of word usage patterns and user relationships across different geographic regions before actual location estimation is needed. These pre-computed distributions capture regional linguistic and social patterns, enabling accurate location estimation even when users do not explicitly mention geographical names in their posts
Solution Approach 2:
Instead of trying to extract location information directly from user text (forward approach), the patent inverts the approach by first establishing what words and relationships are characteristic of each region (base distributions), then using these to infer user locations. This inversion allows location estimation based on implicit behavioral patterns rather than explicit geographical mentions
Data Source
AI summary
A system and method to estimate a location relating to a user who has not filled in information about the location in a profile field in social media such as a microblog. The system and method estimates association between a user in social media and a location includes the steps of acquiring a first content posted to the social media by a first user associated with a first location, determines regional localization of the first content on the basis of the first location, acquires a second content posted to the social media by a second user not associated with a location, determine the degree of a relationship between the first content and the second content, and associating the first location with the second user on the basis of the localization and the degree of the relationship.