Extração de perfis de usuários com e sem deficiência utilizando agrupamentos

Abstract

Nowadays, one can safely affirm that the web has a very important role in our society, allowing broad access to knowledge and simplifying the accomplishment of many different tasks. However, improvements are still needed to ensure that all people, having disabilities or not, enjoy full access to information. Based on this need, assistive technology resources are frequently used, enabling people with disabilities to have autonomy in performing tasks. Due to different interaction characteristics in people with different disabilities, it’s necessary to adapt presented web content. We performed a literature review that found works which present interaction characteristics for users with disabilities and without disabilities, splitting them in distinct profiles based on interaction manners in websites. This work has the purpose of grouping profiles of blind people, people with motor disabilities and people without any disabilities from their way of interacting with web pages. In order to achieve this goal, a system was developed for collecting user information during its interaction with a web page. We then identified the interaction characteristics of each user with disability and created descriptors containing the characteristics of each user profile. Then, two experiments were applied through a screening technique. Soon after, all the information collected was grouped with a non-supervised machine learning technique (K-Means), implemented through the WEKA tool. In the first experiment, some elements were not correctly grouped by the algorithm, leading to corrections in the processing of the collected data. Although the second experiment had incorrectly labeled some elements, its accuracy was higher as measured by the matrix-of-confusion method. The contributions of this work are promising, although more extensive studies involving people with real disabilities and a larger number of participants are necessary to prove their viability in integrating into real adaptive system interfaces. Once we run additional proposed tests and tune system parameters, we can consider its use so, from classification of user profile, real-time web page adaptations could be applied to facilitate navigation.