Methodological proposal to identify the nationality of Twitter users through random-forests

Area of Science:

• Social Network Analysis
• Computational Social Science
• Machine Learning Applications

Background:

• Identifying participants and their contributions in social networks is challenging.
• Understanding user nationality is crucial for accurate opinion analysis, especially in politically sensitive contexts.
• Previous methods for determining user nationality in social networks lacked efficiency and trust.

Purpose of the Study:

• To develop and validate a methodology for identifying the nationality of Twitter users.
• To enhance the accuracy of opinion studies by classifying user nationality.
• To assess the efficiency and trust of the proposed methodology in social network analysis.

Main Methods:

• Utilized machine learning, specifically the Random Forests algorithm, for user classification.
• Employed numerical characteristics derived from user interaction frequencies for model training.
• Focused on a small training sample size for model generation.
• Applied the methodology to analyze Twitter users in three Central American countries.

Main Results:

• The classification model significantly improved the accuracy of nationality identification.
• Average accuracy increased from 77.40% with initial data to 91.60% after applying the classification model, an improvement of 14.20%.
• The methodology demonstrated efficiency and provided a reasonable approach for opinion studies.

Conclusions:

• The proposed methodology offers an effective and efficient approach to determining user nationality on social networks.
• Machine learning, particularly Random Forests, is suitable for classifying users based on interaction data.
• This method enhances the reliability of opinion analysis in social network studies.