Use this link to cite:
https://hdl.handle.net/2183/46408 Desarrollo de un sistema de clasificación de riesgos depresivos en redes sociales mediante análisis de síntomas y grandes modelos de lenguaje
Loading...
Identifiers
Publication date
Authors
Barrios Oliveira, Blanca
Advisors
Other responsabilities
Universidade da Coruña. Facultade de Informática
Journal Title
Bibliographic citation
Type of academic work
Academic degree
Abstract
[Resumen]: Diversos estudios en Natural Language Processing (NLP) e Inteligencia Artificial (IA) han demostrado que la información compartida en redes sociales pueden ser clave para desarrollar sistemas capaces de detectar señales de riesgo depresivo en los usuarios. Cuando estos sistemas incorporan criterios clínicos validados, pueden convertirse en herramientas valiosas para que los profesionales sanitarios realicen intervenciones preventivas aprovechando el gran rendimiento que han demostrado estos modelos en análisis de textos. Este Trabajo de Fin de Grado (TFG) presenta un sistema que, a partir del filtrado de publicaciones con sintomatología según criterios clínicos, utiliza un Large Language Model (LLM) para emitir alertas de riesgo depresivo, considerando además la evolución temporal de los síntomas. El objetivo es proporcionar a los profesionales soluciones que puedan automatizar la revisión de historiales en redes sociales, liberándolos de una tarea tediosa y mejorando la eficiencia y calidad del proceso clínico. La propuesta se evalúa con conocimiento experto sobre el corpus eRisk 2022 de Reddit, demostrando su utilidad para generar explicaciones respaldadas con criterios clínicos y emitir alertas precisas de riesgo.
[Abstract]: Several studies in Natural Language Processing (NLP) and Inteligencia Artificial (IA) have shown that information shared in social networks can be key to developing systems capable of detecting signs of depressive risk in users. When these systems incorporate validated clinical criteria, they can become valuable tools for healthcare professionals to perform preventive interventions by taking advantage of the high performance that these models have demonstrated in text analysis. This Trabajo de Fin de Grado (TFG) presents a system that, based on the filtering of publications with symptomatology according to clinical criteria, uses a Large Language Model (LLM) to issue depressive risk alerts, also considering the temporal evolution of the symptoms. The objective is to provide professionals with solutions that can automate the review of social network histories, freeing them from a tedious task and improving the efficiency and quality of the clinical process. The proposal is evaluated with expert knowledge on Reddit’s eRisk 2022 corpus, demonstrating its usefulness in generating explanations supported with clinical criteria and issuing accurate risk alerts.
[Abstract]: Several studies in Natural Language Processing (NLP) and Inteligencia Artificial (IA) have shown that information shared in social networks can be key to developing systems capable of detecting signs of depressive risk in users. When these systems incorporate validated clinical criteria, they can become valuable tools for healthcare professionals to perform preventive interventions by taking advantage of the high performance that these models have demonstrated in text analysis. This Trabajo de Fin de Grado (TFG) presents a system that, based on the filtering of publications with symptomatology according to clinical criteria, uses a Large Language Model (LLM) to issue depressive risk alerts, also considering the temporal evolution of the symptoms. The objective is to provide professionals with solutions that can automate the review of social network histories, freeing them from a tedious task and improving the efficiency and quality of the clinical process. The proposal is evaluated with expert knowledge on Reddit’s eRisk 2022 corpus, demonstrating its usefulness in generating explanations supported with clinical criteria and issuing accurate risk alerts.
Description
Keywords
Procesamiento de Lenguaje Natural Gran Modelo de Lenguaje Inteligencia Artificial Cadena de Pensamiento Detección de depresión Modelos basados en Transformers Filtrado sintomático Clasificación de riesgos Natural Language Processing Large Language Model Artificial Intelligence Chain-of-Thought Depression Detection Transformer-Based Modelling Symptomatic Filtering Risk Classification
Editor version
Rights
Attribution-NonCommercial-NoDerivatives 4.0 International







