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.
View on arXiv@article{bai2025_2504.06780,
title={ CHIME: A Compressive Framework for Holistic Interest Modeling },
author={ Yong Bai and Rui Xiang and Kaiyuan Li and Yongxiang Tang and Yanhua Cheng and Xialong Liu and Peng Jiang and Kun Gai },
journal={arXiv preprint arXiv:2504.06780},
year={ 2025 }
}