Data and information are the heart of the decision-making process. Information needs require the adoption of innovative technologies and a deeper approach to the object of investigation. The expansion of conversational intelligence over the last decade makes natural language systems a great tool, allowing the user to use voice or text channels to extract and process data. This Thesis proposes an approach for the design of an infrastructure aimed at the extraction, processing, presentation and visualization of structured data, in light of the developments of conversational agents, based on the simplicity of use and on the inference capabilities of the model. The approach exploits a series of elaborations, translations and modeling that allow inexperienced business users to connect a relational database, interact with the data and extract information and knowledge from the results by dialogues. A great contribution to this process is given by large models that represent an important new frontier of conversational intelligence. More specifically, this Thesis focuses on the following contributions: A characterization of the information exploration platform, in terms of functional and non-functional requirements and objectives; A methodology to connect data sources through minimal annotations, explore them through large models, expose the results and manipulate them in order to generate new knowledge thanks to data summarization and data visualization, making possible a deeper exploration; An architecture for a framework that, by exploiting state-of-the-art technologies, supports the distribution of a flexible and modular platform for accessing and processing information; A prototype of the framework that integrates different technologies. The Thesis exploits and reports studies on business users that compared the performance and user satisfaction with regard to this proposal.
I dati e l'informazione sono il cuore del processo decisionale. Le esigenze informative richiedono l'adozione di tecnologie innovative ed un avvicinamento dell'utente all'oggetto di indagine. L'espansione dell'ultimo decennio in materia di intelligenza conversazionale rende i sistemi a linguaggio naturale un ottimo alleato, permettendo all'utente di utilizzare canali vocali o testuali per estrarre e elaborare dati. Questa Tesi propone un approccio per la progettazione di un'infrastruttura finalizzata all'estrazione, elaborazione, presentazione e visualizzazione di dati strutturati, alla luce degli sviluppi sugli agent conversazionali e ponendo come fondamento la semplicità di utilizzo e le capacità inferenziali del modello. L'approccio sfrutta un susseguirsi di elaborazioni, traduzioni e modellazioni che consentono a utenti aziendali inesperti di collegare un database relazionale, interagire con i dati ed estrarre informazioni e conoscenza dai risultati tramite dialoghi. Un grande contributo a questo processo è dato dai large models che rappresentano una nuova importante frontiera dell'intelligenza conversazionale. Questo elaborato verte sui seguenti contributi: Una caratterizzazione della piattaforma per l'esplorazione dell'informazione, in termini di requisiti e obiettivi funzionali e non funzionali; Una metodologia per collegare sorgenti di dati tramite annotazioni minimali, esplorarle tramite large models, esporre i risultati e manipolarli al fine di generare nuova conoscenza grazie a data summarization e data visualization, rendendo possibile una più profonda esplorazione; Un'architettura per un framework che, sfruttando tecnologie all'avanguardia, supporti la distribuzione di una piattaforma flessibile e modulare di accesso all'informazione; Un prototipo del framework che integra diverse tecnologie. La tesi sfrutta e riporta studi su utenti business valutando la loro soddisfazione nei riguardi di tale proposta.
Conversational access to structured knowledge exploiting large models
MANTO, VINCENZO
2022/2023
Abstract
Data and information are the heart of the decision-making process. Information needs require the adoption of innovative technologies and a deeper approach to the object of investigation. The expansion of conversational intelligence over the last decade makes natural language systems a great tool, allowing the user to use voice or text channels to extract and process data. This Thesis proposes an approach for the design of an infrastructure aimed at the extraction, processing, presentation and visualization of structured data, in light of the developments of conversational agents, based on the simplicity of use and on the inference capabilities of the model. The approach exploits a series of elaborations, translations and modeling that allow inexperienced business users to connect a relational database, interact with the data and extract information and knowledge from the results by dialogues. A great contribution to this process is given by large models that represent an important new frontier of conversational intelligence. More specifically, this Thesis focuses on the following contributions: A characterization of the information exploration platform, in terms of functional and non-functional requirements and objectives; A methodology to connect data sources through minimal annotations, explore them through large models, expose the results and manipulate them in order to generate new knowledge thanks to data summarization and data visualization, making possible a deeper exploration; An architecture for a framework that, by exploiting state-of-the-art technologies, supports the distribution of a flexible and modular platform for accessing and processing information; A prototype of the framework that integrates different technologies. The Thesis exploits and reports studies on business users that compared the performance and user satisfaction with regard to this proposal.File | Dimensione | Formato | |
---|---|---|---|
Thesis.pdf
accessibile in internet per tutti
Descrizione: Teso della tesi
Dimensione
3.26 MB
Formato
Adobe PDF
|
3.26 MB | Adobe PDF | Visualizza/Apri |
Executive_Summary.pdf
accessibile in internet per tutti
Descrizione: Executive summary
Dimensione
600.8 kB
Formato
Adobe PDF
|
600.8 kB | 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/10589/208327