Chair: Nicolussi
Brunini, C., Concetti, E., D’Alessandro, G. The Language of Official Statistics Through the Analysis of Data Dissemination Reports
Abstract
| The research investigates the vocabulary of the corpus drawn from 10 different editions of the Annual Report on the State of the Country (RA) by Istat. The goal is to analyse the evolution of the language of official statistics over time, focusing on its syntactic characteristics and communicative value in relation to natural language. The techniques used to process the corpus are those typical of automatic textual data analysis based on statistics. |
Picascia, S., Montanelli, S., Salini, S., Verzillo, S. PubLens: a 2D Visualization Tool for Datasets of Scientific Publications
Abstract
| The growing rate of publication in scientific research demands for advanced tools for effective literature exploration. In this paper, we introduce PubLens, a two-dimensional visualization tool designed to facilitate the interactive exploration of scientific publication datasets, leveraging Large Language Models (LLMs) and dimensionality reduction techniques. As a case study, we demonstrate PubLens on a dataset of 1.3 million research papers from Elsevier Scopus, showcasing its utility in mapping and analyzing the landscape of data science research. |
El Akhal, H., Ben Yahya, A., El Belrhiti El Alaoui, A. Deep Learning vs Deep Hybrid Models: A Performance Evaluation for Olive Diseases Image Classification
Abstract
| Olive trees are highly susceptible to diseases that threaten crop yields and farmers’ incomes. Traditional detection methods are labor-intensive, time-consuming, and require expert knowledge. This study addresses these challenges by evaluating two artificial intelligence (AI) approaches for the automatic detection of olive diseases: (1) Deep Hybrid Models (DHMs), which combine EfficientNetV2B0 for feature extraction with machine learning classifiers (MLP, LR, RF, and SVM), and (2) a Fully Convolutional Neural Network (FCN) based on EfficientNetV2B0. The experiment utilizes a dataset of olive images comprising seven classes. The FCN achieves the highest accuracy (97.07%) but requires significantly more training time. Among the DHMs, EfficientNetV2B0-MLPperforms best (96.54% accuracy), while EfficientNetV2B0-LR trains in just 0.05 minutes. These findings highlight the trade-off between accuracy and computational efficiency. While the FCN excels in performance, its high computational cost limits its applicability in resource-constrained environments. DHMs, particularly EfficientNetV2B0-MLP, offer a balance between accuracy and efficiency, with EfficientNetV2B0-LR being ideal for real-time applications. This study provides valuable insights for selecting optimal models based on specific constraints and needs, contributing to the development of efficient AI-driven disease detection in agriculture. |