GRIAL resources
Permanent URI for this communityhttps://repositorio.grial.eu/handle/123456789/1
Browse
4 results
Search Results
Item Are Textual Recommendations Enough? Guiding Physicians Toward the Design of Machine Learning Pipelines Through a Visual Platform(Springer, 2023-09-05) Vázquez-Ingelmo, A.; García-Holgado, A.; García-Peñalvo, F. J.; Pérez-Sánchez, P.; Antúnez-Muiños, P.; Sánchez-Puente, A.; Vicente-Palacios, V.; Dorado-Díaz, P. I.; Sánchez, P. L.The prevalence of artificial intelligence (AI) in our daily lives is often exaggerated by the media, leading to a positive public perception while overlook-ing potential problems. In the field of medicine, it is crucial to educate future health-care professionals on the advantages and disadvantages of AI and to emphasize the importance of creating fair, ethical, and reproducible models. The KoopaML platform was developed to provide an educational and user-friendly interface for inexperienced users to create AI pipelines. This study analyzes the quantitative and interaction data gathered from a usability test involving physicians from the University Hospital of Salamanca, with the aim of identifying new interaction paradigms to improve the platform’s usability. The results shown that the plat-form is difficult to learn for inexperienced users due to its contents related to AI. Following these results, a set of improvements are proposed for the next version of KoopaML, focusing on reducing the interactions needed to create the pipelines.Item Flexible Heuristics for Supporting RecommendationsWithin an AI Platform Aimed at Non-expert Users(Springer, 2023-05-01) Vázquez-Ingelmo, A.; García-Holgado, A.; García-Peñalvo, F. J.; Andrés-Fraile, E.; Pérez-Sánchez, P.; Antúnez-Muiños, P.; Sánchez-Puente, A.; Vicente-Palacios, V.; Dorado-Díaz, P. I.; Cruz-González, I.; Sánchez, P. L.The use of Machine Learning (ML) to resolve complex tasks has become popular in several contexts. While these approaches are very effective and have many related benefits, they are still very tricky for the general audi-ence. In this sense, expert knowledge is crucial to apply ML algorithms properly and to avoid potential issues. However, in some situations, it is not possible to rely on experts to guide the development of ML pipelines. To tackle this issue, we present an approach to provide customized heuristics and recommendations through a graphical platform to build ML pipelines, namely KoopaML, focused on the medical domain. With this approach, we aim not only at providing an easy way to apply ML for non-expert users, but also at providing a learning experience for them to understand how these methods work.Item Enabling adaptability in web forms based on user characteristics detection through A/B testing and machine learning(IEEE, 2018-02-14) Cruz-Benito, J.; Vázquez-Ingelmo, A.; Sánchez-Prieto, J. C.; Therón, R.; García-Peñalvo, F. J.; Martín-González, M.This paper presents an original study with the aim of improving users' performance in completing large questionnaires through adaptability in web forms. Such adaptability is based on the application of machine-learning procedures and an A/B testing approach. To detect the user preferences, behavior, and the optimal version of the forms for all kinds of users, researchers built predictive models using machine-learning algorithms (trained with data from more than 3000 users who participated previously in the questionnaires), extracting the most relevant factors that describe the models, and clustering the users based on their similar characteristics and these factors. Based on these groups and their performance in the system, the researchers generated heuristic rules between the different versions of the web forms to guide users to the most adequate version (modifying the user interface and user experience) for them. To validate the approach and con rm the improvements, the authors tested these redirection rules on a group of more than 1000 users. The results with this cohort of users were better than those achieved without redirection rules at the initial stage. Besides these promising results, the paper proposes a future study that would enhance the process (or automate it) as well as push its application to other eldsItem Presentation of the paper “Improving success/completion ratio in large surveys: a proposal based on usability and engagement” in HCII 2017(Grupo GRIAL, 2017-07-12) Cruz-Benito, Juan; Therón, R.; García-Peñalvo, Francisco J.; Sánchez-Prieto, J. C.; Vázquez-Ingelmo, A.; Martín-González, M.; Martínez, J. M.This is the presentation of the paper entitled “Improving success/completion ratio in large surveys: a proposal based on usability and engagement” in the Emerging interactive systems for education session at the HCI International 2017 Conference, held in Vancouver, Canada, 9 - 14 July 2017. This paper presents a research focused on improve the success/completion ratio in large surveys. In this case, the large survey is the questionnaire produced by the Spanish Observatory for University Employability and Employment. This questionnaire is composed by about 32 and 60 questions and between 86 and 181 variables to be measured. The research is based on the previous experience of a past questionnaire proposed also by the Observatory composed also by a large amount of questions and variables to be measured (63-92 questions and 176-279 variables). After analysing the target population of the questionnaire (also comparing with the tar-get population of the previous questionnaire) and reviewing the literature, the researchers have designed 11 proposals for changes related to the questionnaire that could improve the users’ completion and success ratios (changes that could improve the users’ trust in the questionnaire, the questionnaire usability and user experience or the users’ engagement to the questionnaire). These changes are planned to be applied in the questionnaire in two main different experiments based on A/B test methodologies that will allow researchers to measure the effect of the changes in different populations and in an incremental way. The proposed changes have been assessed by five experts through an evaluation questionnaire. In this questionnaire, researchers gathered the score of each expert regarding to the pertinence, relevance and clarity of each change proposed. Regarding the results of this evaluation questionnaire, the reviewers fully supported 8 out of the 11 changes proposals, so they could be introduced in the questionnaire with no variation. On the other hand, 3 of the proposed changes or improvements are not fully supported by the experts (they have not received a score in the top first quartile of the 1-7 Likert scale). These changes will not be discarded immediately, because despite they have not received a Q1 score, they received a score within the second quartile of that 1-7 Likert scale, so could be reviewed to be enhanced to fit the OEEU’s context.