This work investigates the application of computational methods for structured and unstructured data. The domains of application are two closely connected fields with the common goal of promoting the stability of the financial system: systemic risk and bank supervision. The work explores different families of models and applies them to different tasks: graphical Gaussian network models to address bank interconnectivity, topic models to monitor bank news and deep learning for text classification. New applications and variants of these models are investigated posing a particular attention on the combined use of textual and structured data. In the penultimate chapter is introduced a sentiment polarity classification tool in Italian, based on deep learning, to simplify future researches relying on sentiment analysis. The different models have proven useful for leveraging numerical (structured) and textual (unstructured) data. Graphical Gaussian Models and Topic models have been adopted for inspection and descriptive tasks while deep learning has been applied more for predictive (classification) problems. Overall, the integration of textual (unstructured) and numerical (structured) information has proven useful for systemic risk and bank supervision related analysis. The integration of textual data with numerical data in fact, has brought either to higher predictive performances or enhanced capability of explaining phenomena and correlating them to other events.
This work investigates the application of computational methods for structured and unstructured data. The domains of application are two closely connected fields with the common goal of promoting the stability of the financial system: systemic risk and bank supervision. The work explores different families of models and applies them to different tasks: graphical Gaussian network models to address bank interconnectivity, topic models to monitor bank news and deep learning for text classification. New applications and variants of these models are investigated posing a particular attention on the combined use of textual and structured data. In the penultimate chapter is introduced a sentiment polarity classification tool in Italian, based on deep learning, to simplify future researches relying on sentiment analysis. The different models have proven useful for leveraging numerical (structured) and textual (unstructured) data. Graphical Gaussian Models and Topic models have been adopted for inspection and descriptive tasks while deep learning has been applied more for predictive (classification) problems. Overall, the integration of textual (unstructured) and numerical (structured) information has proven useful for systemic risk and bank supervision related analysis. The integration of textual data with numerical data in fact, has brought either to higher predictive performances or enhanced capability of explaining phenomena and correlating them to other events.
Six papers on computational methods for the analysis of structured and unstructured data in the economic domain
NICOLA, GIANCARLO
2019-06-10
Abstract
This work investigates the application of computational methods for structured and unstructured data. The domains of application are two closely connected fields with the common goal of promoting the stability of the financial system: systemic risk and bank supervision. The work explores different families of models and applies them to different tasks: graphical Gaussian network models to address bank interconnectivity, topic models to monitor bank news and deep learning for text classification. New applications and variants of these models are investigated posing a particular attention on the combined use of textual and structured data. In the penultimate chapter is introduced a sentiment polarity classification tool in Italian, based on deep learning, to simplify future researches relying on sentiment analysis. The different models have proven useful for leveraging numerical (structured) and textual (unstructured) data. Graphical Gaussian Models and Topic models have been adopted for inspection and descriptive tasks while deep learning has been applied more for predictive (classification) problems. Overall, the integration of textual (unstructured) and numerical (structured) information has proven useful for systemic risk and bank supervision related analysis. The integration of textual data with numerical data in fact, has brought either to higher predictive performances or enhanced capability of explaining phenomena and correlating them to other events.File | Dimensione | Formato | |
---|---|---|---|
PhD thesis Giancarlo Nicola.pdf
accesso aperto
Descrizione: Six papers on computational methods for the analysis of structured and unstructu
Tipologia:
Tesi di dottorato
Dimensione
9.96 MB
Formato
Adobe PDF
|
9.96 MB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.