Search papers, labs, and topics across Lattice.
This paper introduces ANR-DiffRec, a novel framework that enhances discrete diffusion models in recommendation systems by integrating item-based collaborative filtering information. By incorporating an item co-occurrence matrix and an adaptive noise rescheduling mechanism, the method effectively captures both intra-item structural context and inter-item collaborative signals during the denoising process. Experimental results show that ANR-DiffRec significantly outperforms existing state-of-the-art generative recommendation models across multiple benchmarks, highlighting its effectiveness in improving recommendation accuracy.
Integrating item-based collaborative filtering into discrete diffusion models leads to a substantial boost in recommendation accuracy, outperforming existing methods.
Discrete Diffusion Models (DDMs) have recently been introduced to recommendation systems, modeling user history as a token generation process via iterative denoising. However, while effective at capturing user-level sequential patterns, these methods often fail to explicitly integrate item-based collaborative filtering information, a critical component for accurate recommendation. This deficiency manifests in two key aspects: (1) the item representation is often semantic-focused, lacking collaborative priors for diffusion training; and (2) the denoising process employs a uniform noise schedule, treating all tokens indiscriminately and ignoring item-level adaptive structural dependencies. To bridge this gap, we propose ANR-DiffRec, a unified framework designed to encode item-based collaborative structures into discrete diffusion for generative recommendation. First, we explicitly incorporate an item co-occurrence matrix to guide semantic ID generation, providing a structured collaborative prior for discrete diffusion training. Second, we introduce an item-based adaptive noise rescheduling mechanism that dynamically adjusts denoising weights according to both local contextual recoverability and behavior-aware item dependencies. Specifically, the proposed strategy jointly models intra-item structural context and inter-item collaborative signals, enabling structure-aware denoising during diffusion training. Extensive experiments on multiple benchmarks demonstrate that our method consistently outperforms state-of-the-art generative recommendation models. Code: https://github.com/CalmaQi/ANR-DiffRec.
