Listwise Preference Diffusion Optimization for User Behavior Trajectories Prediction

Hongtao Huang (University of New South Wales) · Chengkai Huang (University of New South Wales) · Junda Wu (University of California, San Diego) · Tong Yu (Samsung Research America) · Julian McAuley (UCSD) · Lina Yao (University of New South Wales)

coherent preference generationdenoising steps diffusion modelsindependent-token assumptionlistwise preference diffusion optimizationlistwise ranking likelihoodsmulti-step user behaviorplackett–luce supervisionprobabilistic fidelityreal-world user behavior benchmarkssequential matchstate-of-the-art baselinesstructured preference learningstructured preferencesuser behavior trajectory predictionvariational lower bound

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 longterm 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.