HCHelena de Medeiros Caseli

Helena de Medeiros Caseli

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Helena de Medeiros Caseli has a total of 92 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Engenharia de prompts como técnica pedagógica para leitura crítica da IA em Educação Midiática

    by Alexandra Bujokas de Siqueira, Helena De Medeiros Caseli

    O artigo apresenta resultados parciais de pesquisa de pós-doutorado que articula Processamento de Linguagem Natural e Educação Midiática, testando a engenharia de prompts como dispositivo de leitura crítica da IA generativa na formação inicial de professores. No estudo, exploratório e qualitativo, três formatos de instrução são aplicados a uma notícia em um modelo de pesos abertos, e as saídas são comparadas segundo as funções do enquadramento em Entman (1993). A progressão entre os formatos não é linear e o salto decisivo ocorre com a incorporação de definição teórica ao comando, o que situa a mediação docente como condição, e não como complemento.

  • Open Access

    Explainable AI in Medicine: A Systematic Literature Mapping Study

    by Sadeeq Olalekan Bello, Giovanna Castro, Luiza Barioto, Renato Moraes Silva, Helena Caseli, João Agostinho Machado-Neto, Aline Villavicencio, Tiago Almeida, Ricardo Cerri

    Explainable artificial intelligence (XAI) has been proposed to address concerns regarding AI system opacity in clinical applications. However, a systematic understanding of XAI in healthcare remains limited. This study presents a systematic mapping of XAI methods, clinical areas, and evaluation approaches across medical domains. A search across five databases (2020-2025) filtered to Q1/Q2 journals yielded 383 studies. Three methods were dominant: SHAP (54.0%), LIME (18.3%), and Grad-CAM (17.8%). Research concentrated on oncology (25.1%), neurology (14.6%), cardiology (10.2%), and pulmonology (6.8%), with diagnosis as the primary application (49.9%). These four domains represented 56.7% of all studies. Method selection closely relates to data type: SHAP accounts for 60.1% of structured data applications, while Grad-CAM appears in 85% of imaging studies. Evaluation remains mostly technical, with 63.4% relying on quantitative metrics, 8.4% incorporating human assessment, and 12.3% conducting external validation. Research shows a concentration of methods (three methods account for 90.6% of studies), domains, and evaluation practices (91.6% lack human assessment). The reviewed studies demonstrate technical capability but provide limited evidence of clinical utility. Bridging the gap between computational validation and clinical assessment requires the comparative evaluation of dominant methods, expansion into underrepresented domains, development of multimodal approaches, and validation protocols testing explanation utility in clinical contexts. Without systematic clinical evaluation, the extent to which current practices address healthcare needs remains unclear.

  • LLM-Based Solution Applied to Explore Healthcare Datasets

    by Fernando Rezende Zagatti, André Gomes Regino, Filipe Loyola Lopes, Gilson Yuuji Shimizu, Rodrigo Bonacin, Daniel Lucrédio, Helena de Medeiros Caseli

  • Open Access

    A Machine Learning Classification Model for Identifying College Students with Depression Based on Digital Phenotyping

    by Evandro Y. A. Ribeiro, Franco E. Garcia, Conrado dos S. Alves Saud, Helena de M. Caseli, Vivian G. Motti, Taís Bleicher, Jair B. Neto, Heloisa C. Figueiredo Frizzo, Larissa C. Martini, Luciano de O. Neris, et al.

    Depression is a serious global mental health illness that causes significant suffering to the individual and social impairment in their lives. Compared to the general population, depression shows a higher prevalence among college students. With recent advancements in digital phenotyping data analysis to infer depressive symptoms, machine learning (ML) techniques have been increasingly employed to indicate behaviors related to potential depressive profiles (PDP). However, despite the growing body of work on ML usage to detect depression, few studies have focused on data preprocessing approaches to handle missing values in datasets that go beyond common data imputation. In this study, we conducted a series of experiments to evaluate the combination of data preprocessing methods and ML algorithms for effectively classifying PDP and non-PDP students using data from the Amive project. The primary challenges were implementing a data processing workflow to address missing values and class imbalance, common issues in digital phenotyping datasets, and selecting algorithms capable of handling such data. The experimental results showed promising outcomes, with individual classification models, including Random Forest, XGBoost, and SVM(rbf), achieving accuracies of 77%, 75%, and 76%, respectively. The best performance was obtained by training on datasets that went through outlier filtering, specifically removing rows with four or more missing values. This combination of data preprocessing approaches and ML algorithms resulted in a Random Forest classification model with the best performance ranging between 77% of accuracy and with mean errors metrics of AUC and MCC above 0.5.

  • Automated Machine Learning in medical research: A systematic literature mapping study

    by Giovanna A. Castro, Luiza G. Barioto, Yu H. Cao, Renato M. Silva, Helena M. Caseli, João A. Machado-Neto, Ricardo Cerri, Aline Villavicencio, Tiago A. Almeida


  • Open Access

    Investigating the Relationship Between Text Vectorization Cosine Similarity and Classification Performance

    by Fernando Rezende Zagatti, Gilson Yuuji Shimizu, Daniel Lucrédio, Helena de Medeiros Caseli

    In recent years, optimizing classification pipelines has become increasingly critical due to the growing volume of textual data and the computational challenges associated with exhaustive hyperparameter tuning. This paper proposes a similarity-based approach for selecting the most promising vectorization configurations–specifically, Bag of Words (BoW), Term Frequency-Inverse Document Frequency (TF-IDF), and Word2Vec–by analyzing the average cosine similarity of the generated vectors; by preselecting configurations that yield more diverse textual representations, our method relies on the hypothesis that increased diversity in text representations enhances the discriminative capacity of classification models. Experimental evaluations conducted on five different datasets demonstrate that the similarity-based approach achieves accuracy and F1-Score results very close to those obtained via exhaustive search, with notable reductions in processing time; furthermore, correlation analyses reveal a strong inverse relationship between vector similarity and model performance for BoW and TF-IDF, and a moderate relationship for Word2Vec. These findings validate the efficacy of the proposed method as a practical alternative to hyperparameter selection in vectorization pipelines, offering significant benefits for applications where exhaustive exploration is unfeasible.


  • Unsupervised Statistical Keyword Extraction Pipeline: Is LLM All You Need?

    by Fernando Rezende Zagatti, Daniel Lucrédio, Helena de Medeiros Caseli


  • Evaluating Sentiment Quantification Methods in Brazilian Portuguese Corpora

    by Lucas Nildaimon dos Santos Silva, Diego Furtado Silva, Helena de Medeiros Caseli

  • Open Access

    CÂMARAS DE ECO POLÍTICAS DURANTE OS ATOS ANTIDEMOCRÁTICOS

    by William Tsuyoshi Matsuda, Helena de Medeiros Caseli, Alan Demétrius Baria Valejo, Sylvia Iasulaitis

    A formação de câmaras de eco (echo chambers) consiste em um fenômeno no qual os usuários são frequentemente expostos a pontos de vista que reforçam suas crenças, dificultando a exposição a perspectivas divergentes. Enquanto sabe-se do surgimento e da disseminação das câmaras de eco que se formam em comunidades online, a identificação automática e compreensão deste fenômeno surge como uma importante ferramenta para auxiliar a investigação das suas implicações na esfera política e na democracia. O objetivo deste artigo é a investigação, implementação e avaliação de métodos para a detecção de câmaras de eco em redes sociais digitais durante os Atos Antidemocráticos de 8 de Janeiro de 2023. A investigação desse fenômeno foi realizada por meio da combinação de modelagem de comunidades baseada em redes, da análise de posicionamento político e da análise de sentimentos e emoções. Para tanto, foi criado um modelo computacional com a finalidade de identificar câmaras de eco presentes em um corpus de tweets coletado entre os dias 8 e 10 de janeiro de 2023. Os dados obtidos demonstram uma nítida separação que reflete uma polarização significativa na rede social, com grupos distintos de usuários discutindo principalmente dentro de suas próprias bolhas ideológicas, evidenciando a existência de câmaras de eco. A análise dos tópicos extraídos das comunidades revela padrões distintos de conversação entre os grupos com diferentes polaridades políticas, sendo as discussões nessas comunidades centradas na defesa ou oposição aos atos e às ações do governo e das instituições que buscaram contê-los. Foi possível identificar, ainda, a prevalência de emoções como raiva e aborrecimento, já que a troca de acusações e a retórica inflamada contribuem para a expressão predominante de emoções negativas.

  • Open Access

    Significant changes in occupations among college students: a study carried out on a correlated online social network profile

    by Sérgio Antônio Mendes Reche, Jair Borges Barbosa Neto, Helena de Medeiros Caseli, Heloísa Cristina Figueiredo Frizzo

    This study examines changes in occupational performance as observed in posts by university students on a social network. Depression is prevalent in Brazil, particularly among students, impacting academic and daily functioning. Using CrowdTangle, anonymous Facebook posts were analyzed following American Occupational Therapy Association guidelines. 301 posts revealed occupational changes across seven areas defined by the association. Depressive symptoms hinder academic performance, exacerbating symptoms. Academic stress, loneliness, stigma, loss of leisure interest, and socioeconomic factors contribute to distress. Social engagement and support benefit mental health, while time constraints detrimentally affect academics, especially disadvantaged students. The study underscores the link between occupational changes, academic decline, and depression, proposing six relevant subcategories within the Education domain.