MMQ: Multimodal Mixture-of-Quantization Tokenization for Semantic ID Generation and User Behavioral Adaptation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Xu, Yi, Zhang, Moyu, Li, Chenxuan, Liao, Zhihao, Xing, Haibo, Deng, Hao, Hu, Jinxin, Zhang, Yu, Zeng, Xiaoyi, Zhang, Jing
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918364951085056
author Xu, Yi
Zhang, Moyu
Li, Chenxuan
Liao, Zhihao
Xing, Haibo
Deng, Hao
Hu, Jinxin
Zhang, Yu
Zeng, Xiaoyi
Zhang, Jing
author_facet Xu, Yi
Zhang, Moyu
Li, Chenxuan
Liao, Zhihao
Xing, Haibo
Deng, Hao
Hu, Jinxin
Zhang, Yu
Zeng, Xiaoyi
Zhang, Jing
contents Recommender systems traditionally represent items using unique identifiers (ItemIDs), but this approach struggles with large, dynamic item corpora and sparse long-tail data, limiting scalability and generalization. Semantic IDs, derived from multimodal content such as text and images, offer a promising alternative by mapping items into a shared semantic space, enabling knowledge transfer and improving recommendations for new or rare items. However, existing methods face two key challenges: (1) balancing cross-modal synergy with modality-specific uniqueness, and (2) bridging the semantic-behavioral gap, where semantic representations may misalign with actual user preferences. To address these challenges, we propose Multimodal Mixture-of-Quantization (MMQ), a two-stage framework that trains a novel multimodal tokenizer. First, a shared-specific tokenizer leverages a multi-expert architecture with modality-specific and modality-shared experts, using orthogonal regularization to capture comprehensive multimodal information. Second, behavior-aware fine-tuning dynamically adapts semantic IDs to downstream recommendation objectives while preserving modality information through a multimodal reconstruction loss. Extensive offline experiments and online A/B tests demonstrate that MMQ effectively unifies multimodal synergy, specificity, and behavioral adaptation, providing a scalable and versatile solution for both generative retrieval and discriminative ranking tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2508_15281
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle MMQ: Multimodal Mixture-of-Quantization Tokenization for Semantic ID Generation and User Behavioral Adaptation
Xu, Yi
Zhang, Moyu
Li, Chenxuan
Liao, Zhihao
Xing, Haibo
Deng, Hao
Hu, Jinxin
Zhang, Yu
Zeng, Xiaoyi
Zhang, Jing
Information Retrieval
Machine Learning
Recommender systems traditionally represent items using unique identifiers (ItemIDs), but this approach struggles with large, dynamic item corpora and sparse long-tail data, limiting scalability and generalization. Semantic IDs, derived from multimodal content such as text and images, offer a promising alternative by mapping items into a shared semantic space, enabling knowledge transfer and improving recommendations for new or rare items. However, existing methods face two key challenges: (1) balancing cross-modal synergy with modality-specific uniqueness, and (2) bridging the semantic-behavioral gap, where semantic representations may misalign with actual user preferences. To address these challenges, we propose Multimodal Mixture-of-Quantization (MMQ), a two-stage framework that trains a novel multimodal tokenizer. First, a shared-specific tokenizer leverages a multi-expert architecture with modality-specific and modality-shared experts, using orthogonal regularization to capture comprehensive multimodal information. Second, behavior-aware fine-tuning dynamically adapts semantic IDs to downstream recommendation objectives while preserving modality information through a multimodal reconstruction loss. Extensive offline experiments and online A/B tests demonstrate that MMQ effectively unifies multimodal synergy, specificity, and behavioral adaptation, providing a scalable and versatile solution for both generative retrieval and discriminative ranking tasks.
title MMQ: Multimodal Mixture-of-Quantization Tokenization for Semantic ID Generation and User Behavioral Adaptation
topic Information Retrieval
Machine Learning
url https://arxiv.org/abs/2508.15281