Large language models (LLMs) have emerged as powerful tools in the realm of conversational agents, offering capabilities to understand, generate, and respond to human language with remarkable sophistication. One significant challenge facing conversational agents is accurately understanding context based on spatial or temporal cues, which requires sophisticated mechanisms to interpret and adapt to the dynamic nuances of space and time within conversations. This thesis presents an innovative protocol aimed at overcoming the challenge of spatial and temporal context awareness in conversational agents. It outlines a systematic approach to empower LLMs with the capability to effectively interpret and respond to such contextual cues. The methodology is exemplified through the implementation of Schindly, a virtual assistant integrated into a real elevator environment. Subsequently, a real-world case study within a business building was conducted over a two-week period, providing empirical evidence of the protocol's efficacy in enhancing conversational agent performance. This research not only contributes to the advancement of conversational agent technology but also showcases its practical implications in real-world scenarios, paving the way for enhanced human-computer interactions across various spatial and temporal contexts.
I modelli linguistici di grandi dimensioni (LLMs) sono emersi come strumenti potenti nel campo degli agenti conversazionali, offrendo capacità di comprendere, generare e rispondere al linguaggio umano con una notevole sofisticatezza. Una sfida significativa che affrontano gli agenti conversazionali è comprendere accuratamente il contesto basato su indizi spaziali o temporali, il che richiede meccanismi sofisticati per interpretare e adattarsi alle sfumature dinamiche dello spazio e del tempo all'interno delle conversazioni. Questa tesi presenta un protocollo innovativo mirato a superare la sfida della consapevolezza del contesto spaziale e temporale negli agenti conversazionali. Esso delineare un approccio sistematico per dotare i LLMs della capacità di interpretare ed rispondere in modo efficace a tali indizi contestuali. La metodologia è esemplificata attraverso l'implementazione di Schindly, un assistente virtuale integrato in un vero ambiente di ascensore. Successivamente, è stata condotta una case study nel mondo reale all'interno di un edificio aziendale per un periodo di due settimane, fornendo evidenze empiriche dell'efficacia del protocollo nel migliorare le prestazioni degli agenti conversazionali. Questa ricerca non solo contribuisce al progresso della tecnologia degli agenti conversazionali, ma mostra anche le sue implicazioni pratiche in scenari reali, aprendo la strada per interazioni umano-computer migliorate in vari contesti spaziali e temporali.
Large Language Models as smart space aware social conversational agents
Sonnino, Nicolò
2022/2023
Abstract
Large language models (LLMs) have emerged as powerful tools in the realm of conversational agents, offering capabilities to understand, generate, and respond to human language with remarkable sophistication. One significant challenge facing conversational agents is accurately understanding context based on spatial or temporal cues, which requires sophisticated mechanisms to interpret and adapt to the dynamic nuances of space and time within conversations. This thesis presents an innovative protocol aimed at overcoming the challenge of spatial and temporal context awareness in conversational agents. It outlines a systematic approach to empower LLMs with the capability to effectively interpret and respond to such contextual cues. The methodology is exemplified through the implementation of Schindly, a virtual assistant integrated into a real elevator environment. Subsequently, a real-world case study within a business building was conducted over a two-week period, providing empirical evidence of the protocol's efficacy in enhancing conversational agent performance. This research not only contributes to the advancement of conversational agent technology but also showcases its practical implications in real-world scenarios, paving the way for enhanced human-computer interactions across various spatial and temporal contexts.File | Dimensione | Formato | |
---|---|---|---|
2024_04_Sonnino_Executive_Summary.pdf
accessibile in internet per tutti
Dimensione
697.69 kB
Formato
Adobe PDF
|
697.69 kB | Adobe PDF | Visualizza/Apri |
2024_04_Sonnino_Thesis.pdf
accessibile in internet per tutti
Dimensione
19.45 MB
Formato
Adobe PDF
|
19.45 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/10589/219872