Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: He, Eric, Gupta, Akash, Liusie, Adian, Raina, Vatsal, Molenda, Piotr, Chabra, Shirom, Raina, Vyas
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2510.12014
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866912645871828992
author He, Eric
Gupta, Akash
Liusie, Adian
Raina, Vatsal
Molenda, Piotr
Chabra, Shirom
Raina, Vyas
author_facet He, Eric
Gupta, Akash
Liusie, Adian
Raina, Vatsal
Molenda, Piotr
Chabra, Shirom
Raina, Vyas
contents Text--image retrieval is necessary for applications such as product recommendation. Embedding-based approaches like CLIP enable efficient large-scale retrieval via vector similarity search, but they are primarily trained on literal caption-like text--image pairs and often fail to capture abstract or persona-driven attributes common in product recommendation applications (e.g., ``a gift for a mother who loves gardening''). In contrast, state-of-the-art vision--language models (vLLMs) can align text with images in a flexible manner, but their limited context window prevents them from directly handling retrieval over large catalogs. We propose a framework that distills the preference rankings of a powerful vLLM into an embedding-based system, transferring its nuanced alignment abilities while maintaining the inference-time scalability of an embedding-based approach. Experiments on persona-driven product recommendation tasks demonstrate that our method significantly outperforms existing embedding-based baselines, providing an efficient solution for personalized text--image retrieval.
format Preprint
id arxiv_https___arxiv_org_abs_2510_12014
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Embedding the Teacher: Distilling vLLM Preferences for Scalable Image Retrieval
He, Eric
Gupta, Akash
Liusie, Adian
Raina, Vatsal
Molenda, Piotr
Chabra, Shirom
Raina, Vyas
Information Retrieval
Machine Learning
Text--image retrieval is necessary for applications such as product recommendation. Embedding-based approaches like CLIP enable efficient large-scale retrieval via vector similarity search, but they are primarily trained on literal caption-like text--image pairs and often fail to capture abstract or persona-driven attributes common in product recommendation applications (e.g., ``a gift for a mother who loves gardening''). In contrast, state-of-the-art vision--language models (vLLMs) can align text with images in a flexible manner, but their limited context window prevents them from directly handling retrieval over large catalogs. We propose a framework that distills the preference rankings of a powerful vLLM into an embedding-based system, transferring its nuanced alignment abilities while maintaining the inference-time scalability of an embedding-based approach. Experiments on persona-driven product recommendation tasks demonstrate that our method significantly outperforms existing embedding-based baselines, providing an efficient solution for personalized text--image retrieval.
title Embedding the Teacher: Distilling vLLM Preferences for Scalable Image Retrieval
topic Information Retrieval
Machine Learning
url https://arxiv.org/abs/2510.12014