Content-Based Web Classifier System for Dementia Definitions Using Natural Language Processing

This article presents a methodology designed to determine the taxonomy of dementia definitions using a natural language processing model. Data was collected from 30 healthcare professionals through in-depth interviews to develop the model, and fixed lists of words were created based on these definitions. The model proposes three taxonomic approaches for defining dementia: biomedical, psychosocial, and colloquial. Two similarity metrics were used to evaluate the level of belonging of a definition to each of these approaches: the Jaccard index and the vector cosine. These metrics provide percentages that indicate the degree of belonging of a definition to the approaches proposed in the taxonomy. As part of the functional validation process of the model, experiments were performed using new definitions of dementia drawn from the scientific literature. In total, 30 experiments were carried out, which included twenty definitions of a biomedical approach, five of a psychosocial nature, and five with a daily approach. The results show that the model has an accuracy of 100% for the Jaccard index and 93.1% for the vector model. Although these results indicate a high level of performance in content-based classifiers, these results could be improved by adding more tokens to the fixed lists through interviews with health professionals from other branches of medicine. This study could give clues on how to improve scientific communication and disseminate dementia research that is not focused on the biomedical model.