Percepción de la calidad del servicio en el sistema de salud mediante análisis de opinión en bases de datos disponibles en la web.
Loading...
Date
2025
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad de Concepción
Abstract
El presente trabajo tuvo como objetivo analizar la percepción de la calidad del servicio de salud en Chile a partir de opiniones de usuarios recopiladas en plataformas públicas en línea, utilizando técnicas de Procesamiento de Lenguaje Natural (PLN) y aprendizaje automático. La motivación principal radica en la falta de métodos automatizados de análisis de opinión en el contexto chileno, en contraste con la creciente disponibilidad de comentarios públicos en plataformas digitales. A través de web crawling automatizado, se recolectaron más de 2.000 opiniones de seis hospitales públicos y seis clínicas privadas, seleccionadas en base a criterios de tamaño poblacional y distribución geográfica.
Posteriormente, los textos fueron sometidos a un exhaustivo proceso de preprocesamiento, incluyendo la eliminación de acentos, normalización, tokenización y vectorización, para garantizar su correcta interpretación por los modelos analíticos. Para la clasificación de sentimientos, se implementaron tres algoritmos supervisados: Naive Bayes, Máquina de Vectores de Soporte (SVM) y una Red Neuronal Simple, utilizando la calificación numérica adjunta a cada opinión como criterio de etiquetado. Se decidió eliminar la clase neutra debido a su escasa representación, transformando el problema en una clasificación binaria entre sentimientos positivos y negativos. Los resultados mostraron que Naive Bayes obtuvo el mejor desempeño en hospitales, mientras que SVM lideró en clínicas, alineándose en parte con lo reportado en la literatura especializada.
En la segunda etapa, se aplicaron métodos de modelado de tópicos mediante aprendizaje no supervisado, utilizando Latent Dirichlet Allocation (LDA) y Factorización de Matrices No Negativas (FMN). Ambos enfoques permitieron identificar patrones temáticos recurrentes en las opiniones, tales como tiempos de espera, trato del personal y calidad de la atención, siendo FMN el que logró una mejor definición y segmentación de los tópicos, especialmente en los subconjuntos de opiniones negativas.
En conjunto, los resultados obtenidos evidencian que el uso de técnicas de PLN y machine learning puede automatizar de manera efectiva el análisis de la percepción ciudadana sobre el sistema de salud, proporcionando información valiosa para la mejora continua del servicio. Asimismo, el algoritmo desarrollado presenta potencial para ser implementado en sistemas de monitoreo en tiempo real, permitiendo a las instituciones de salud detectar tendencias en la opinión de los usuarios de manera temprana y oportuna, favoreciendo la toma de decisiones basada en evidencia.
This work aimed to analyze the perception of healthcare service quality in Chile based on user opinions collected from public online platforms, utilizing Natural Language Processing (NLP) techniques and machine learning. The primary motivation lies in the lack of automated opinion analysis methods in the Chilean context, despite the increasing availability of public comments on digital platforms. Through automated web crawling, more than 2.000 opinions were collected from six public hospitals and six private clinics, selected based on population size and geographical distribution criteria. The texts underwent an extensive preprocessing phase, including accent removal, normalization, tokenization, and vectorization, to ensure their correct interpretation by the analytical models. For sentiment classification, three supervised algorithms were implemented: Naive Bayes, Support Vector Machine (SVM), and a Simple Neural Network, using the numerical rating associated with each opinion as a labeling criterion. Due to its low representation, the neutral class was eliminated, transforming the problem into a binary classification between positive and negative sentiments. The results showed that Naive Bayes achieved the best performance in hospitals, while SVM outperformed the other models in clinics, partially aligning with findings reported in the literature. In the second stage, topic modeling techniques based on unsupervised learning were applied, using Latent Dirichlet Allocation (LDA) and Non-negative Matrix Factorization (NMF). Both approaches enabled the identification of recurring thematic patterns in the opinions, such as waiting times, staff treatment, and quality of care, with NMF achieving a better definition and segmentation of topics, especially among negative opinions. Overall, the results demonstrate that applying NLP and machine learning techniques can effectively automate the analysis of public perception of the healthcare system, providing valuable information for continuous service improvement. Furthermore, the developed algorithm shows potential for real-time monitoring applications, allowing healthcare institutions to detect emerging trends in user opinions in a timely manner and to support evidence-based decision-making.
This work aimed to analyze the perception of healthcare service quality in Chile based on user opinions collected from public online platforms, utilizing Natural Language Processing (NLP) techniques and machine learning. The primary motivation lies in the lack of automated opinion analysis methods in the Chilean context, despite the increasing availability of public comments on digital platforms. Through automated web crawling, more than 2.000 opinions were collected from six public hospitals and six private clinics, selected based on population size and geographical distribution criteria. The texts underwent an extensive preprocessing phase, including accent removal, normalization, tokenization, and vectorization, to ensure their correct interpretation by the analytical models. For sentiment classification, three supervised algorithms were implemented: Naive Bayes, Support Vector Machine (SVM), and a Simple Neural Network, using the numerical rating associated with each opinion as a labeling criterion. Due to its low representation, the neutral class was eliminated, transforming the problem into a binary classification between positive and negative sentiments. The results showed that Naive Bayes achieved the best performance in hospitals, while SVM outperformed the other models in clinics, partially aligning with findings reported in the literature. In the second stage, topic modeling techniques based on unsupervised learning were applied, using Latent Dirichlet Allocation (LDA) and Non-negative Matrix Factorization (NMF). Both approaches enabled the identification of recurring thematic patterns in the opinions, such as waiting times, staff treatment, and quality of care, with NMF achieving a better definition and segmentation of topics, especially among negative opinions. Overall, the results demonstrate that applying NLP and machine learning techniques can effectively automate the analysis of public perception of the healthcare system, providing valuable information for continuous service improvement. Furthermore, the developed algorithm shows potential for real-time monitoring applications, allowing healthcare institutions to detect emerging trends in user opinions in a timely manner and to support evidence-based decision-making.
Description
Tesis presentada para optar al título de Ingeniero/a Civil Biomédico.
Keywords
Sistemas de salud, Opinión pública, Bases de datos, Lingüística computacional, Machine learning