Epidemic outbreaks constitute a major and persistent threat to global health, as demonstrated by the COVID-19 pandemic, which exposed critical vulnerabilities in healthcare infrastructures, economic systems, and societal organization, as well as substantial gaps in preparedness for public health emergencies. In this context, Genomic Surveillance has emerged as a key scientific instrument for supporting evidence-based decision-making by enabling the systematic collection, monitoring, and analysis of pathogen genomic data at national and international scales. This thesis investigates how Genomic Surveillance-related themes are represented across two heterogeneous information sources, namely scientific publications and news articles, through the enhancement of an established topic modeling architecture designed for large-scale textual corpora. First, we adapt and improve this architecture to support the comparative analysis of domain-specific topics extracted from distinct corpora. Second, we applies Bidirectional Topic Matching (BTM) to quantify semantic correspondences between topics generated by different models through similarity-based metrics. Third, we extend the use of BTM beyond pairwise model comparison by introducing a graph-based representation that enables the simultaneous comparison of multiple topic models. Within this graph structure, community-detection algorithms are employed to identify macro-topics, defined as coherent groups of topics sharing a common contextual and semantic scope. Finally, the thesis integrates multiple visualization strategies, including graph representations, temporal trends, and heat maps, to provide an interpretable overview of the resulting macro-topics and to support the systematic validation of the extracted thematic structures. Overall, the proposed approach offers a scalable methodological framework for comparing topic structures across heterogeneous textual sources and for analysing the evolution and communication of Genomic Surveillance-related knowledge.
Le epidemie rappresentano una minaccia rilevante e persistente per la salute globale, come dimostrato dalla pandemia di COVID-19, che ha messo in luce vulnerabilità critiche nelle infrastrutture sanitarie, nei sistemi economici e nell’organizzazione sociale, oltre a significative lacune nella preparazione alle emergenze di sanità pubblica. In questo contesto, la Sorveglianza Genomica è emersa come uno strumento scientifico fondamentale per supportare processi decisionali basati sull’evidenza, consentendo la raccolta, il monitoraggio e l’analisi sistematica dei dati genomici dei patogeni su scala nazionale e internazionale. Questa tesi indaga il modo in cui i temi legati alla Sorveglianza Genomica sono rappresentati in due sorgenti di dati eterogenee, ovvero pubblicazioni scientifiche e articoli di giornale, attraverso il miglioramento di un’architettura consolidata di topic modeling progettata per grandi corpora testuali. In primo luogo, abbiamo adattato e migliorato tale architettura per facilitare l’analisi comparativa di topic specifici estratti da corpora distinti. Successivamente, abbiamo applicato il Bidirectional Topic Matching (BTM) per quantificare le corrispondenze semantiche tra topic generati da modelli differenti mediante metriche basate sulla similarità. Infine, abbiamo esteso l’utilizzo del BTM oltre il confronto a coppie tra modelli, introducendo una rappresentazione basata su grafi che consente il confronto simultaneo di più modelli. Alla struttura a grafo sono stati applicati algoritmi di community detection per identificare macro-topic, definiti come gruppi coerenti di topic che condividono un comune ambito contestuale e semantico. Infine, la tesi integra diverse tecniche di visualizzazione: tra cui grafi, serie temporali e heat map, al fine di fornire una panoramica interpretabile dei macro-topic risultanti e di supportarne la validazione. Nel complesso, l’approccio proposto offre un framework metodologico scalabile per confrontare strutture tematiche tra fonti testuali eterogenee e per analizzare l’evoluzione e la comunicazione della conoscenza relativa alla Sorveglianza Genomica.
Comparing genomic surveillance topics across news and scientific literature in the last 25 years
Banfi, Stefano Alessandro
2025/2026
Abstract
Epidemic outbreaks constitute a major and persistent threat to global health, as demonstrated by the COVID-19 pandemic, which exposed critical vulnerabilities in healthcare infrastructures, economic systems, and societal organization, as well as substantial gaps in preparedness for public health emergencies. In this context, Genomic Surveillance has emerged as a key scientific instrument for supporting evidence-based decision-making by enabling the systematic collection, monitoring, and analysis of pathogen genomic data at national and international scales. This thesis investigates how Genomic Surveillance-related themes are represented across two heterogeneous information sources, namely scientific publications and news articles, through the enhancement of an established topic modeling architecture designed for large-scale textual corpora. First, we adapt and improve this architecture to support the comparative analysis of domain-specific topics extracted from distinct corpora. Second, we applies Bidirectional Topic Matching (BTM) to quantify semantic correspondences between topics generated by different models through similarity-based metrics. Third, we extend the use of BTM beyond pairwise model comparison by introducing a graph-based representation that enables the simultaneous comparison of multiple topic models. Within this graph structure, community-detection algorithms are employed to identify macro-topics, defined as coherent groups of topics sharing a common contextual and semantic scope. Finally, the thesis integrates multiple visualization strategies, including graph representations, temporal trends, and heat maps, to provide an interpretable overview of the resulting macro-topics and to support the systematic validation of the extracted thematic structures. Overall, the proposed approach offers a scalable methodological framework for comparing topic structures across heterogeneous textual sources and for analysing the evolution and communication of Genomic Surveillance-related knowledge.| File | Dimensione | Formato | |
|---|---|---|---|
|
2026_07_Banfi_Tesi.pdf
accessibile in internet per tutti a partire dal 01/07/2029
Descrizione: Testo della tesi
Dimensione
13.38 MB
Formato
Adobe PDF
|
13.38 MB | Adobe PDF | Visualizza/Apri |
|
2026_07_Banfi_executive_summary.pdf
accessibile in internet per tutti a partire dal 01/07/2029
Descrizione: executive summary
Dimensione
2.52 MB
Formato
Adobe PDF
|
2.52 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/260208