Abstract

Abstract
E-commerce has become an essential aspect of modern life, providing consumers worldwide with convenience and accessibility. However, the high volume of short and noisy product descriptions in text streams of massive e-commerce platforms translates into an increased number of clusters, presenting challenges for standard model-based stream clustering algorithms. This is the case of a dataset extracted from the Brazilian NF-e Project containing electronic invoice product descriptions, including many product clusters. While LDA-based clustering methods have shown to be crucial, they have been mainly evaluated on datasets with few clusters. We propose the Topic Model with Contextual Outlier Handling (TMCOH) method to overcome this limitation. This method combines the Dirichlet Process, specific word representation, and contextual outlier detection techniques to recycle identified outliers aiming to integrate them into appropriate clusters later on. The experimental results for our case study demonstrate the effectiveness of TMCOH when compared to state-of-the-art methods and its potential for application to text clustering in large datasets.

Abstract
Digital transformation plays a significant role in modernizing and improving the efficiency of ports around the world. However, digitalization also brings a set of challenges that ports must face. They have to respond to several unique challenges because of the complexity of their operations and the varying demands of stakeholders. This study seeks to identify and summarize the challenges of digital transformation processes in ports. For this purpose, the World Ports Sustainability Program database was used. The findings revealed 74 digitalization initiatives carried out by ports, which makes it possible to recognize 7 dimensions and 32 sub-dimensions of challenges to the digital transformation process. Among the identified dimensions are port infrastructure, the interconnection between various systems, the port organization model, regulation, security and privacy, market evolution, and the establishment of partnerships to implement these projects. The results of this study are relevant to mitigate the risks of the digitalization process in ports and respond to market needs that demand greater transparency and visibility of their operations.

Abstract
In a photovoltaic power plant (PVPP), the DC-AC converter (inverter) is one of the components most prone to faults. Even though they are key equipment in such installations, their fault detection techniques are not as much explored as PV module-related issues, for instance. In that sense, this paper is motivated to find novel tools for detection focused on the inverter, employing machine learning (ML) algorithms trained using a hybrid dataset. The hybrid dataset is composed of real and synthetic data for fault-free and faulty conditions. A dataset is built based on fault-free data from the PVPP and faulty data generated by a digital twin (DT). The combination DT and ML is employed using a Clarke/space vector representation of the inverter electrical variables, thus resulting in a novel feature engineering method to extract the most relevant features that can properly represent the operating condition of the PVPP. The solution that was developed can classify multiple operation conditions of the inverter with high accuracy.

Abstract
Enterprise Architecture (EA) is a coherent set of principles, methods, and models that express the structure of an enterprise and its IT landscape. EA mining uses data mining techniques to automate EA modelling tasks. Ontologies help to define concepts and the relationships among these concepts to describe a domain of interest This work presents an extensible ontology for EA mining focused on extracting architectural models that use logs from an API gateway as the data source. The proposed ontology was developed using the OntoUML language to ensure its quality and avoid anti-patterns through ontology rule checks. Then, a hypothesized scenario using data structures close to the real is used to simulate the ontology application and validate its theoretical feasibility as well as its contribution to the Enterprise Architecture Management field.

Abstract
The application of machine learning (ML) algorithms to electronic health records (EHR) data allows the achievement of data-driven insights on various clinical problems and the development of clinical decision support (CDS) systems to improve patient care. However, data governance and privacy barriers hinder the use of data from multiple sources, especially in the medical field due to the sensitivity of data. Federated learning (FL) is an attractive data privacy-preserving solution in this context by enabling the training of ML models with data from multiple sources without any data sharing, using distributed remotely hosted datasets. The Secur-e-Health project aims at developing a solution in terms of CDS tools encompassing FL predictive models and recommendation systems. This tool may be especially useful in Pediatrics due to the increasing demands on Pediatric services, and the current scarcity of ML applications in this field compared to adult care. Herein we provide a description of the technical solution proposed in this project for three specific pediatric clinical problems: childhood obesity management, pilonidal cyst post-surgical care and retinography imaging analysis.