This thesis presents an advanced search engine capable of processing multimodal queries (text, image, or both), using embeddings generated with BLIP-2 and indexed in vector databases. Two approaches, PGVector and Milvus, are compared by evaluating retrieval effectiveness on proprietary datasets. The analysis shows that the multimodal approach improves the relevance of the results, especially in complex scenarios.
Questa tesi presenta un motore di ricerca avanzato in grado di elaborare query multimodali (testo, immagine o entrambi), utilizzando embedding generati con BLIP-2 e indicizzati in database vettoriali. Due approcci, PGVector e Milvus, vengono confrontati valutando l’efficacia del recupero su dataset proprietari. L’analisi mostra che l’approccio multimodale migliora la rilevanza dei risultati, soprattutto in scenari complessi.
MULTIMODAL ADVANCED SEARCH ON VECTOR DATABASES
FIORETTI, MARIA LUCIA
2024/2025
Abstract
This thesis presents an advanced search engine capable of processing multimodal queries (text, image, or both), using embeddings generated with BLIP-2 and indexed in vector databases. Two approaches, PGVector and Milvus, are compared by evaluating retrieval effectiveness on proprietary datasets. The analysis shows that the multimodal approach improves the relevance of the results, especially in complex scenarios.| File | Dimensione | Formato | |
|---|---|---|---|
|
Tesi_Fioretti.pdf
File consultabile da postazioni biblioteca
Descrizione: This thesis presents an advanced search engine capable of processing multimodal queries (text, image, or both), using embeddings generated with BLIP-2 and indexed in vector databases. Two approaches, PGVector and Milvus, are compared by evaluating retriev
Dimensione
4.14 MB
Formato
Adobe PDF
|
4.14 MB | Adobe PDF |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14253/64622