Time-Aware Diffusion based on Preference Disentanglement for Generative Recommendation

2026-06-01 · Source: Takara TLDR - Daily AI Papers · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Recommendation Systems · Depth: Expert, medium

Summary

TDPM, a novel Generative Recommender (GR) framework, addresses a critical limitation in existing diffusion-based GRs by integrating time-evolving user preferences into the diffusion process on semantic ID (SID) tokens. While prior diffusion GRs apply uniformly across historical interactions, TDPM disentangles user preference into a long-term consistent "period preference" and a "point preference" triggered by recent events. This approach accounts for the non-stationary, multifaceted nature of user preferences over time. Extensive experiments on three public real-world datasets demonstrate TDPM's significant superiority, achieving average improvements of up to 29.21% in HR@20 and 25.45% in NDCG@20 over leading baselines. An ablation study further confirms the necessity of its time-aware token diffusion mechanism.

Key takeaway

For machine learning engineers developing generative recommendation systems, you should consider incorporating time-aware diffusion mechanisms. Your models can achieve substantial performance gains by disentangling user preferences into long-term "period" and recent "point" components, as demonstrated by TDPM's 29.21% HR@20 improvement. This approach directly addresses the non-stationary nature of user interests, leading to more accurate and relevant recommendations.

Key insights

Time-aware diffusion and preference disentanglement significantly enhance generative recommender performance by modeling evolving user interests.

Principles

• User preferences are non-stationary and time-evolving.
• Disentangling preference into period and point components improves modeling.
• Time-aware diffusion is crucial for generative recommenders.

Method

TDPM designs time-aware diffusion on SID tokens by disentangling user preference into a long-term "period preference" and a "point preference" from recent events, integrating these into the diffusion process.

In practice

• Implement time-aware mechanisms in diffusion models.
• Decompose user preferences into stable and event-driven components.
• Utilize semantic ID tokens for generative recommendation.

Topics

• Generative Recommenders
• Diffusion Models
• Time-Aware Modeling
• Preference Disentanglement
• Semantic IDs
• Recommendation Systems

Code references

• LinkMingzhe/DualFashion

Best for: Research Scientist, AI Scientist, Machine Learning Engineer

Related on AIssential

• ProbRes: Volatility Learning for Probabilistic Time-Series Forecasting — ProbRes is a post-hoc calibration method that learns volatility dynamics for robust probabilistic time series forecasting in heteroskedastic data.
• Bridging the Last Mile of Time Series Forecasting with LLM Agents — LLM agents can bridge the gap between statistical time series forecasts and business-ready predictions by integrating weakly structured contextual information.
• Physically-Constrained Mamba-SDE for Remaining Useful Life Prediction under Irregular Observations — Embedding physical priors into continuous-time latent dynamics improves Remaining Useful Life prediction under irregular observations.
• VLBM: Variational Latent Basis Modeling for OOD Robust Multivariate Time Series Forecasting — VLBM improves multivariate time series forecasting by explicitly separating stable dynamics from OOD deviations using a latent basis model.
• Why Do Time Series Models Need Long Context Windows? — Long context windows in time series forecasting primarily reduce uncertainty in identifying the data-generating process, crucial for optimal predictions.
• E4GEN: Event-level Explainable Extreme-Enhanced Time-series Generation — E4GEN is an explainable diffusion framework that precisely controls extreme event generation in time series without disrupting overall fidelity.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.