Generative AI has seen a notable increase in use over the past few years, changing various fields from natural language processing to creative arts. This change is mainly due to the introduction of the Transformer model in 2017, which improved the handling of sequen- tial data. Recently, Large Language Models (LLMs) like OpenAI’s GPT-4o have gained popularity, showcasing their potential to automate and improve many applications. Ad- vertising, particularly digital advertising, has evolved with the rise of the Internet and social networks. Companies now use various media like search engines and web banners. However, managing advertising budgets across multiple platforms can be challenging and inefficient. To help solve this problem, MINT introduced Advertising Resource Manage- ment (ARM), a software platform designed to consolidate advertising operations. ARM’s platform uses real-time data analysis and predictive insights to help businesses better manage their digital advertising efforts. Despite its benefits, the complexity of the cur- rent web application can make it difficult for users to engage with the platform. This thesis aims to improve user interaction with the ARM platform by introducing an AI assistant powered by LLMs. The assistant allows users to ask questions in natural lan- guage, which are then converted into SQL queries to retrieve relevant data. The study evaluates various LLMs, including those from OpenAI, Google, and Anthropic, for their performance in this task. A new safety model is also proposed to detect potentially harm- ful requests, adding an extra layer of security. the accuracy of generated SQL queries. The AI assistant architecture is designed to be scalable and adaptable, allowing for easy updates and improvements. Results show significant improvements in query accuracy and user engagement, making data retrieval more straightforward and efficient. This thesis demonstrates the practical benefits of LLMs in digital advertising, suggesting potential for future improvements in AI-driven user assistance technologies.
L'AI generativa ha registrato un notevole aumento nell'utilizzo negli ultimi anni, cambiando vari campi dalla elaborazione del linguaggio naturale alle arti creative. Questo cambiamento è dovuto principalmente all'introduzione del modello Transformer nel 2017, che ha migliorato la gestione dei dati sequenziali. Recentemente, i Large Language Model (LLM), come GPT-4o di OpenAI, hanno guadagnato popolarità, dimostrando il loro potenziale per automatizzare e migliorare molte applicazioni. L'advertising, in particolare il digital advertising, si è evoluto con l'ascesa di Internet e dei social network. Le aziende ora utilizzano vari media come i motori di ricerca e i banner web. Tuttavia, gestire i budget pubblicitari su più piattaforme può essere impegnativo e inefficiente. Per aiutare a risolvere questo problema, MINT ha introdotto l'Advertising Resource Management (ARM), una piattaforma software progettata per consolidare le operazioni pubblicitarie. La piattaforma ARM utilizza l'analisi dei dati in tempo reale e approfondimenti predittivi per aiutare le aziende a gestire meglio i loro sforzi di pubblicità digitale. Nonostante i suoi vantaggi, la complessità dell'attuale applicazione web può rendere difficile per gli utenti interagire con la piattaforma. Questa tesi mira a migliorare l'interazione degli utenti con la piattaforma ARM introducendo un assistente AI alimentato da LLM. L'assistente consente agli utenti di fare domande in linguaggio naturale, che vengono poi convertite in query SQL per recuperare i dati pertinenti. Lo studio valuta vari LLMs, tra cui quelli di OpenAI, Google e Anthropic, per la loro performance in questo compito. Viene anche proposto un nuovo modello di sicurezza per rilevare richieste potenzialmente dannose, aggiungendo un ulteriore livello di sicurezza. I risultati mostrano miglioramenti significativi nell'accuratezza delle query e nell'engagement degli utenti, rendendo il recupero dei dati più semplice ed efficiente. Questa tesi dimostra i benefici pratici degli LLM nella pubblicità digitale, suggerendo potenziali miglioramenti futuri nelle tecnologie di assistenza agli utenti basate sull'intelligenza artificiale.
Large Language Models to Query Your Data: Retrieving Advertising Industry User's Data Using Natural Language
Rossi, Alessandro
2023/2024
Abstract
Generative AI has seen a notable increase in use over the past few years, changing various fields from natural language processing to creative arts. This change is mainly due to the introduction of the Transformer model in 2017, which improved the handling of sequen- tial data. Recently, Large Language Models (LLMs) like OpenAI’s GPT-4o have gained popularity, showcasing their potential to automate and improve many applications. Ad- vertising, particularly digital advertising, has evolved with the rise of the Internet and social networks. Companies now use various media like search engines and web banners. However, managing advertising budgets across multiple platforms can be challenging and inefficient. To help solve this problem, MINT introduced Advertising Resource Manage- ment (ARM), a software platform designed to consolidate advertising operations. ARM’s platform uses real-time data analysis and predictive insights to help businesses better manage their digital advertising efforts. Despite its benefits, the complexity of the cur- rent web application can make it difficult for users to engage with the platform. This thesis aims to improve user interaction with the ARM platform by introducing an AI assistant powered by LLMs. The assistant allows users to ask questions in natural lan- guage, which are then converted into SQL queries to retrieve relevant data. The study evaluates various LLMs, including those from OpenAI, Google, and Anthropic, for their performance in this task. A new safety model is also proposed to detect potentially harm- ful requests, adding an extra layer of security. the accuracy of generated SQL queries. The AI assistant architecture is designed to be scalable and adaptable, allowing for easy updates and improvements. Results show significant improvements in query accuracy and user engagement, making data retrieval more straightforward and efficient. This thesis demonstrates the practical benefits of LLMs in digital advertising, suggesting potential for future improvements in AI-driven user assistance technologies.File | Dimensione | Formato | |
---|---|---|---|
Large_Language_Models_to_Query_Your_Data_Retrieving_Advertising_Industry_Users__Data_Using_Natural_Language.pdf
accessibile in internet solo dagli utenti autorizzati
Dimensione
2.15 MB
Formato
Adobe PDF
|
2.15 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/10589/226333