Design acessível orientado por dados: classificação automática de descrições comportamentais do personaut via PLN.

Abstract

Developing applications for autistic audiences remains a challenge for lay designers and developers. ProAut is a process that supports these professionals in designing application interfaces for autistic individuals. GuideAut, in turn, is a platform created to support the use of ProAut, bringing together all the necessary artifacts to follow its process, including PersonAut, a behavioral model used to generate profiles of autistic users and assist in building empathy. In this context, this work proposes to use Natural Language Processing (NLP) techniques to automatically classify behavioral descriptions provided by family members/professionals and organize them into the categories that make up PersonAut. The main objective is to develop a model capable of identifying, from natural texts, dimensions such as Calming Activities, Stressful Activities, Social and Family Aspects, Stereotypies/Manias, and Technological Aspects, allowing support in the generation of PersonAut within the GuideAut platform. To achieve this, the standard data science workflow was adopted, including literature review, requirements definition, data collection and preprocessing, selection and training of NLP models, and preparation for integration with GuideAut. The main contribution is expected to be that the results will assist professionals using the platform by automating the construction of structured behavioral profiles and indirectly contributing to the development of more accessible and suitable solutions for the needs of autistic individuals.