This project optimizes Recommender Systems (DeepCoNN, NARRE, DIRECT) for Item Ranking through a dual intervention. Training the Error-Based models with the Bayesian Personalized Ranking (BPR) loss demonstrates significant superiority in ranking metrics over MSE optimization. Furthermore, an evolutionary framework integrating LLM Embeddings is validated. This pipeline leverages the offline pre-computation of rich embeddings to feed specialized ranking models, ensuring high efficiency and suggestion accuracy.
Il progetto ottimizza i Recommender Systems (DeepCoNN, NARRE, DIRECT) per l'Item Ranking mediante un duplice intervento. L'addestramento dei modelli Error-Based con la loss Bayesian Personalized Ranking (BPR) dimostra una superiorità significativa nelle metriche di ranking rispetto all'ottimizzazione MSE. Inoltre, viene validato un framework evolutivo che integra gli LLM Embeddings. Tale pipeline sfrutta il pre-calcolo offline di embeddings ricchi per alimentare modelli di ranking specializzati, garantendo elevata efficienza e accuratezza del suggerimento.
EVOLUZIONE DEI RECOMMENDER SYSTEMS ATTRAVERSO UN APPROCCIO RANKING-BASED E LLM EMBEDDINGS
MARZANO, ANTONIO
2024/2025
Abstract
This project optimizes Recommender Systems (DeepCoNN, NARRE, DIRECT) for Item Ranking through a dual intervention. Training the Error-Based models with the Bayesian Personalized Ranking (BPR) loss demonstrates significant superiority in ranking metrics over MSE optimization. Furthermore, an evolutionary framework integrating LLM Embeddings is validated. This pipeline leverages the offline pre-computation of rich embeddings to feed specialized ranking models, ensuring high efficiency and suggestion accuracy.| File | Dimensione | Formato | |
|---|---|---|---|
|
Antonio Marzano Tesi.pdf
non disponibili
Dimensione
1.59 MB
Formato
Adobe PDF
|
1.59 MB | Adobe PDF |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14253/64088