Towards A Tri-View Diffusion Framework for Recommendation DOI: XXXXXXX.XXXXXXX Conference: 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining; August 9–13, 2026; Jeju, Korea ISBN: 978-1-4503-XXXX-X/2018/06 CCS: Information systems Recommender systems

Ximing Chen, Pui Ieng Lei, Yijun Sheng, Yanyan Liu, Zhiguo Gong · 2025

Read Towards A Tri-View Diffusion Framework for Recommendation DOI: XXXXXXX.XXXXXXX Conference: 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining; August 9–13, 2026; Jeju, Korea ISBN: 978-1-4503-XXXX-X/2018/06 CCS: Information systems Recommender systems by Ximing Chen, Pui Ieng Lei, Yijun Sheng, Yanyan Liu, Zhiguo Gong (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-20122.txt.

Description

Chen, Lei, Sheng, Liu, and Gong rebuild recommender diffusion models on thermodynamic grounds. Their 2025 paper proposes TV-Diff, a tri-view diffusion framework. Diffusion models have grown popular for recommendation tasks. Prior work adopted their goals without much checking. The team first asks whether those goals suffice.

Viewed through heat physics, these models maximize energy alone. Classic systems instead work by reducing entropy. Neither view alone captures the full problem. TV-Diff maximizes Helmholtz free energy, combining both. A custom denoiser preserves anisotropy in user-item links. That unevenness measures cross-links on the two-sided graph. A sampling process called AR-GSP supplies hard negatives. Acceptance-rejection sampling ensures those rejects are informative. A Gumbel softmax step adapts the sampling over time. Experiments show clear gains in accuracy and efficiency. Theory backs the free-energy formulation as well. Sparse suggestion data motivates the whole design.

Catalog record

CREATORXiming Chen, Pui Ieng Lei, Yijun Sheng, Yanyan Liu, Zhiguo Gong
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE69 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-20122

Browse the full library · About the Tower of Babel