Predicting road freight prices is a challenging task influenced by multiple factors. Understanding which variables have the greatest impact is essential for building more accurate models, and consequently for enhancing the competitiveness of Brazilian soybeans in the global market. This study aims to evaluate the influence of different exogenous variables on soybean freight prices and to analyze how this influence varies across different distance ranges. To achieve this, a combination of machine learning techniques was applied to a comprehensive dataset containing information on freight costs, regional characteristics, production, fuel prices, storage, and commercialization. The results indicate that distance is the most significant variable in determining freight costs, directly reflecting operational expenses such as fuel consumption and labor costs. Additionally, macroeconomic factors such as the exchange rate and export volume play a crucial role, highlighting the global context of Brazil’s soybean exports. Stratified analysis by distance ranges reveals distinct patterns; short-distance freight is predominantly related to domestic markets, while medium- and long-distance freight are strongly linked to export logistics.

While depressive linguistic expressions have been extensively studied in traditional clinical contexts, there has been comparatively little attention devoted to modeling how both depressed and non-depressed individuals express their symptoms on social networking sites as a holistic thematic process. This study addresses this gap by examining how depression is expressed linguistically on social networking sites using various topic modeling techniques, including an innovative methodology based on LLMs. We use datasets in the Brazilian Portuguese language gathered from Instagram, Reddit, and X. Our evaluation reveals that while common themes related to depression emerge across different social networking sites, each platform's unique characteristics influence the thematic content. Reddit discussions focus on symptomatology, Instagram on travel and positive emotions, and Twitter on everyday life and media. The LLM-based approach produced more interpretable topics with a higher embedding-based coherence metric, whereas traditional methods often resulted in noisy and less internally coherent topics. This research contributes to a deeper understanding of the holistic online expressions of depression and highlights the potential of advanced topic modeling techniques to reveal subtle aspects of mental health discussions online.

Risk analysis is a determining factor for every project, aiming to assess existing risks with the goal of minimizing financial and human losses. A well-established method for evaluating risks is the risk matrix, which comprises two dimensions: the probability and severity of events. Current machine learning approaches for constructing risk matrices require hundreds or thousands of manually labeled examples by domain experts and typically address only one dimension of the matrix. This work introduces RMGNN (Risk Matrix Generation with Graph Neural Networks), a semi-supervised method that constructs complete risk matrices using only five percent labeled data. RMGNN builds a unified graph structure where events are connected based on textual similarity, then extracts both dimensions from this structure: occurrence probability is estimated through graph centrality measures, while event severity is predicted through semi-supervised label propagation using a graph neural network. For risk managers, this approach reduces expert annotation costs by ninety-five percent and enables building customized risk matrices in days rather than weeks. The method also supports cross-domain knowledge transfer, allowing organizations to leverage labeled incidents from one domain to assess risks in related domains with limited historical data. Experimental evaluation on over ten thousand consumer product incidents demonstrates that RMGNN achieves eighty-one percent accuracy and seventy-four percent F1-score, significantly outperforming traditional semi-supervised baselines while requiring substantially less labeled training data.

Este trabalho apresenta o desenvolvimento de um sistema para monitoramento de sintomas depressivos a partir de publicações em redes sociais, com foco na plataforma Reddit. Por meio da combinação de técnicas de Processamento de Linguagem Natural e modelos de aprendizado de máquina, as publicações foram classificadas quanto ao potencial indício de depressão, sendo complementadas por análises léxicas e temáticas. A aplicação web desenvolvida permite visualização estatística e monitoramento em tempo real, com funcionalidades voltadas ao apoio de profissionais de saúde mental. Os resultados obtidos evidenciam uma correlação significativa entre o conteúdo das comunidades analisadas e a presença de sinais depressivos, sugerindo o potencial do sistema como ferramenta auxiliar na triagem precoce de sintomas. Questões éticas relacionadas à privacidade e à acurácia das classificações também são discutidas.