Polarización estructural en grafos y redes sociales: un estudio temporal sobre OpenAI y DeepSeek en X
Loading...
Official URL
Full text at PDC
Publication date
2025
Authors
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citation
Abstract
Este Trabajo de Fin de Master se centra en el análisis de la polarización estructural en grafos, tomando como el caso de estudio de la interacción de usuarios en la plataforma X (anteriormente Twitter) en torno a los modelos de lenguaje OpenAI y DeepSeek. Para ello, se recopilaron datos manualmente a lo largo de once semanas mediante técnicas de web scraping, identificando usuarios que mencionan o responden a cuentas relacionadas con estos modelos. A partir de dicha información, se construyó una red no dirigida de usuarios y se aplicaron métodos de análisis de grafos, polarización basada en fuzzy-sets: JDJ y análisis temporal. Una vez definida una partición forzada de la red en dos comunidades principales,
se estimaron los grados de pertenencia ideológica de cada nodo usando la proporción e vecinos en cada comunidad. Esta información permitió calcular el índice JDJ, una medida difusa de polarización ideológica desarrollada en trabajos previos. El análisis presentó niveles de polarización estructural altos y persistentes, con momentos puntuales de convergencia entre comunidades, reflejados tanto en la visualización de grafos como en la evolución del índice JDJ. Los resultados muestran que la red estudiada evoluciona hacia una estructura
de tipo small-world, con comunidades cada vez más cohesivas y a veces separadas. Asimismo, se observa que los valores del índice JDJ reflejan con precisión los cambios estructurales y discursivos detectados en la red. El estudio aporta una contribución metodológica al combinar técnicas de análisis estructural, lógica difusa y visualización de datos en el estudio de procesos de polarización ideológica en contextos reales.
This thesis focuses on the analysis of ideological polarization in graphs, using user interactions on the platform X (Twitter) as a case study, specifically regarding the discourse surrounding OpenAI and DeepSeek language models. Data was manually collected over eleven weeks using web scraping techniques, identifying users who mentioned or replied to accounts related to these models. Based on this information, an undirected user network was constructed, and methods of graph analysis, fuzzy-set based polarization (JDJ index), and temporal analysis were applied. After defining a forced partition of the network into two main communities, each node’s ideological affiliation degree was estimated based on the proportion of its neighbors belonging to each community. This information enabled the calculation of the JDJ index, a fuzzy measure of ideological polarization developed in previous works. The analysis revealed consistently high levels of structural polarization, with occasional moments of convergence between communities, observable both in the graph visualizations and in the evolution of the JDJ index. The results indicate that the network evolves towards a small-world structure, with increasingly cohesive and distinct communities. Furthermore, the JDJ index accurately captures both structural and discursive changes within the network. This study offers a methodological contribution by combining structural network analysis, fuzzy logic, and data visualization techniques to examine ideological polarization processes in real world contexts.
This thesis focuses on the analysis of ideological polarization in graphs, using user interactions on the platform X (Twitter) as a case study, specifically regarding the discourse surrounding OpenAI and DeepSeek language models. Data was manually collected over eleven weeks using web scraping techniques, identifying users who mentioned or replied to accounts related to these models. Based on this information, an undirected user network was constructed, and methods of graph analysis, fuzzy-set based polarization (JDJ index), and temporal analysis were applied. After defining a forced partition of the network into two main communities, each node’s ideological affiliation degree was estimated based on the proportion of its neighbors belonging to each community. This information enabled the calculation of the JDJ index, a fuzzy measure of ideological polarization developed in previous works. The analysis revealed consistently high levels of structural polarization, with occasional moments of convergence between communities, observable both in the graph visualizations and in the evolution of the JDJ index. The results indicate that the network evolves towards a small-world structure, with increasingly cohesive and distinct communities. Furthermore, the JDJ index accurately captures both structural and discursive changes within the network. This study offers a methodological contribution by combining structural network analysis, fuzzy logic, and data visualization techniques to examine ideological polarization processes in real world contexts.