Listwise Preference Diffusion Optimization for User Behavior Trajectories Prediction

2025-11-01

AI Summary

Conventional sequential recommendation methods struggle to capture global, list-level dependencies within user behavior sequences, leading to inaccurate multi-step trajectory predictions. Method: This paper introduces the novel task of “user behavioral trajectory prediction” and pioneers the application of diffusion models to structured preference learning. We propose a list-level preference diffusion optimization framework, using the Plackett–Luce distribution as a supervision signal and deriving a variational lower bound on list-wise ranking likelihood—thereby relaxing the independence assumptions inherent in autoregressive approaches. Contribution/Results: We design a dual-metric evaluation protocol—SeqMatch and PPL—and demonstrate significant improvements over state-of-the-art methods across multiple real-world benchmarks. Our results empirically validate the effectiveness of modeling long-horizon, structured user preferences and establish diffusion models as a new paradigm for list-level sequential recommendation.

Technology Category

• Machine Learning: Learning Preferences or Rankings
• Search and Optimization: Learning to Search
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Content-based information diffusion

Abstract

Forecasting multi-step user behavior trajectories requires reasoning over structured preferences across future actions, a challenge overlooked by traditional sequential recommendation. This problem is critical for applications such as personalized commerce and adaptive content delivery, where anticipating a user's complete action sequence enhances both satisfaction and business outcomes. We identify an essential limitation of existing paradigms: their inability to capture global, listwise dependencies among sequence items. To address this, we formulate User Behavior Trajectory Prediction (UBTP) as a new task setting that explicitly models long-term user preferences. We introduce Listwise Preference Diffusion Optimization (LPDO), a diffusion-based training framework that directly optimizes structured preferences over entire item sequences. LPDO incorporates a Plackett-Luce supervision signal and derives a tight variational lower bound aligned with listwise ranking likelihoods, enabling coherent preference generation across denoising steps and overcoming the independent-token assumption of prior diffusion methods. To rigorously evaluate multi-step prediction quality, we propose the task-specific metric Sequential Match (SeqMatch), which measures exact trajectory agreement, and adopt Perplexity (PPL), which assesses probabilistic fidelity. Extensive experiments on real-world user behavior benchmarks demonstrate that LPDO consistently outperforms state-of-the-art baselines, establishing a new benchmark for structured preference learning with diffusion models.

Problem

• Predicting multi-step user behavior trajectories with structured preferences
• Overcoming limitations in capturing global listwise dependencies among items
• Optimizing coherent preference generation across entire action sequences

Innovation

• Diffusion-based training optimizes structured preferences over sequences
• Plackett-Luce supervision aligns with listwise ranking likelihoods
• Tight variational bound enables coherent preference generation steps