Fecha
2025
Editor/a
Director/a
Tutor/a
Coordinador/a
Prologuista
Revisor/a
Ilustrador/a
Derechos de acceso
info:eu-repo/semantics/openAccess
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Nacional de Educación a Distancia (UNED)
Citas
0 citas en 
0 citas en 
Resumen
Este trabajo aborda el desarrollo de un sistema de análisis y visualización temática a partir del Servicio de Quejas y Sugerencias del Ayuntamiento de Zaragoza, con el objetivo de extraer las principales preocupaciones ciudadanas mediante técnicas de procesamiento del lenguaje natural, en particular modelado de tópicos y clustering. Estas técnicas permiten estructurar y comprender grandes volúmenes de texto no estructurado.
Se ha diseñado una metodología basada en varias fases: desde la recopilación y preprocesamiento de datos, hasta la evaluación cuantitativa y cualitativa de los distintos algoritmos de modelado de temas y clustering seleccionados, pasando por un post-clustering. Para finalizar, se ha realizado una integración de una visualización interactiva en la sede electrónica del Ayuntamiento. Se ha realizado una completa y diversa selección de algoritmos para abarcar los diferentes tipos de modelos asistentes: tradicionales como LDA, específicos para textos cortos como GSDMM y basados en redes neuronales como BERTopic. En el caso del clustering, se utilizaron algoritmos clásicos basados en particiones (K-means) y jerárquicos (aglomerativo), combinados con dos representaciones de características complementarias: TF-IDF, centrada en el contenido literal, y Word2Vec, orientada a la semántica. La evaluación cuantitativa se ha basado en métricas extrínsecas (ARI, NMI y Bcubed), tomando como referencia (ground truth) el código de servicio asociado a cada queja o sugerencia. Esta evaluación se ha complementado con una evaluación humana, centrada en dos aspectos: (1) la calidad de los tópicos, evaluando las palabras más representativas de los 20 más grandes; y (2) la coherencia entre los tópicos y los documentos más representativos asociados. Asimismo, se ha abordado el desafío de manejar un número elevado de categorías iniciales (61), proponiendo un proceso de post-clustering que ha facilitado la estructuración y simplificación temática para su representación visual. Para la visualización final, se han valorado las diferentes opciones existentes para modelado de tópicos. Se ha optado por Treemap, dada su capacidad para mostrar jerarquías temáticas de forma intuitiva y enlazar palabras clave con los documentos más representativos. Esta opción supera limitaciones de otras herramientas como pyLDAvis o wordclouds, al reforzar la interpretabilidad y el contexto semántico de los temas detectados. En conjunto, este trabajo constituye una contribución innovadora en el ámbito de la administración pública, al aplicar técnicas avanzadas de PLN a un servicio real de interacción ciudadana, contribuyendo así a la transparencia, la mejora de la calidad de los servicios municipales y la toma de decisiones basada en evidencia.
This work addresses the development of a thematic analysis and visualization system based on the Complaints and Suggestions Service of the Zaragoza City Council, with the aim of identifying the main citizen concerns using natural language processing (NLP) techniques, particularly topic modeling and clustering. These techniques enable the structuring and understanding of large volumes of unstructured text. A multi-phase methodology has been designed, encompassing data collection and preprocessing, the quantitative and qualitative evaluation of various topic modeling and clustering algorithms, and a post-clustering stage. Finally, an interactive visualization has been integrated into the City Council’s electronic headquarters. A comprehensive and diverse selection of algorithms was carried out to cover different types of models.traditional models such as LDA were included, along with models specific to short texts like GS- DMM, and neural network-based models such as BERTopic. For clustering, classical partition-based algorithms (K-means) and hierarchical methods (agglomerative clustering) were used, in combination with two complementary feature representations: TF-IDF, focused on literal content, and Word2Vec, aimed at capturing semantic meaning. Quantitative evaluation was based on extrinsic metrics (ARI, NMI, and Bcubed), using the service code associated with each complaint or suggestion as the ground truth. This was complemented by human evaluation focused on two aspects: (1) topic quality, by assessing the most representative words of the 20 largest topics; and (2) topic-document coherence, by evaluating the relevance of the five most representative documents for each topic. The challenge of handling a high number of initial categories (61) was also addressed through a post-clustering process, which facilitated the thematic structuring and simplification needed for visualization. For the final visualization, various available topic modeling tools were assessed. Treemap was selected due to its ability to intuitively represent thematic hierarchies and link keywords to the most representative documents. This approach overcomes the limitations of other tools such as pyLDAvis or word clouds, by enhancing interpretability and the semantic context of the identified topics. Overall, this work represents an innovative contribution to the field of public administration by applying advanced NLP techniques to a real citizen interaction service, thereby promoting transparency, improving the quality of municipal services, and supporting evidence-based decision-making.
This work addresses the development of a thematic analysis and visualization system based on the Complaints and Suggestions Service of the Zaragoza City Council, with the aim of identifying the main citizen concerns using natural language processing (NLP) techniques, particularly topic modeling and clustering. These techniques enable the structuring and understanding of large volumes of unstructured text. A multi-phase methodology has been designed, encompassing data collection and preprocessing, the quantitative and qualitative evaluation of various topic modeling and clustering algorithms, and a post-clustering stage. Finally, an interactive visualization has been integrated into the City Council’s electronic headquarters. A comprehensive and diverse selection of algorithms was carried out to cover different types of models.traditional models such as LDA were included, along with models specific to short texts like GS- DMM, and neural network-based models such as BERTopic. For clustering, classical partition-based algorithms (K-means) and hierarchical methods (agglomerative clustering) were used, in combination with two complementary feature representations: TF-IDF, focused on literal content, and Word2Vec, aimed at capturing semantic meaning. Quantitative evaluation was based on extrinsic metrics (ARI, NMI, and Bcubed), using the service code associated with each complaint or suggestion as the ground truth. This was complemented by human evaluation focused on two aspects: (1) topic quality, by assessing the most representative words of the 20 largest topics; and (2) topic-document coherence, by evaluating the relevance of the five most representative documents for each topic. The challenge of handling a high number of initial categories (61) was also addressed through a post-clustering process, which facilitated the thematic structuring and simplification needed for visualization. For the final visualization, various available topic modeling tools were assessed. Treemap was selected due to its ability to intuitively represent thematic hierarchies and link keywords to the most representative documents. This approach overcomes the limitations of other tools such as pyLDAvis or word clouds, by enhancing interpretability and the semantic context of the identified topics. Overall, this work represents an innovative contribution to the field of public administration by applying advanced NLP techniques to a real citizen interaction service, thereby promoting transparency, improving the quality of municipal services, and supporting evidence-based decision-making.
Descripción
Categorías UNESCO
Palabras clave
topic models, clustering, visualización, quejas y sugerencias, topic models, clustering, visualisation, complaints and suggestions
Citación
Notivol Bezares, Ruben. Trabajo fin de Máster: "Extracción y visualización de temáticas del Servicio online de quejas y sugerencias del Ayuntamiento de Zaragoza". Universidad Nacional de Educación a Distancia (UNED) 2025
Centro
E.T.S. de Ingeniería Informática
Departamento
Inteligencia Artificial