Saved in:
Bibliographic Details
Main Authors: Bai, Yong, Xiang, Rui, Li, Kaiyuan, Tang, Yongxiang, Cheng, Yanhua, Liu, Xialong, Jiang, Peng, Gai, Kun
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2504.06780
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916681232678912
author Bai, Yong
Xiang, Rui
Li, Kaiyuan
Tang, Yongxiang
Cheng, Yanhua
Liu, Xialong
Jiang, Peng
Gai, Kun
author_facet Bai, Yong
Xiang, Rui
Li, Kaiyuan
Tang, Yongxiang
Cheng, Yanhua
Liu, Xialong
Jiang, Peng
Gai, Kun
contents Modeling holistic user interests is important for improving recommendation systems but is challenged by high computational cost and difficulty in handling diverse information with full behavior context. Existing search-based methods might lose critical signals during behavior selection. To overcome these limitations, we propose CHIME: A Compressive Framework for Holistic Interest Modeling. It uses adapted large language models to encode complete user behaviors with heterogeneous inputs. We introduce multi-granular contrastive learning objectives to capture both persistent and transient interest patterns and apply residual vector quantization to generate compact embeddings. CHIME demonstrates superior ranking performance across diverse datasets, establishing a robust solution for scalable holistic interest modeling in recommendation systems.
format Preprint
id arxiv_https___arxiv_org_abs_2504_06780
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle CHIME: A Compressive Framework for Holistic Interest Modeling
Bai, Yong
Xiang, Rui
Li, Kaiyuan
Tang, Yongxiang
Cheng, Yanhua
Liu, Xialong
Jiang, Peng
Gai, Kun
Information Retrieval
Modeling holistic user interests is important for improving recommendation systems but is challenged by high computational cost and difficulty in handling diverse information with full behavior context. Existing search-based methods might lose critical signals during behavior selection. To overcome these limitations, we propose CHIME: A Compressive Framework for Holistic Interest Modeling. It uses adapted large language models to encode complete user behaviors with heterogeneous inputs. We introduce multi-granular contrastive learning objectives to capture both persistent and transient interest patterns and apply residual vector quantization to generate compact embeddings. CHIME demonstrates superior ranking performance across diverse datasets, establishing a robust solution for scalable holistic interest modeling in recommendation systems.
title CHIME: A Compressive Framework for Holistic Interest Modeling
topic Information Retrieval
url https://arxiv.org/abs/2504.06780