Action-On-Item Preference Flow: A Shared Event Schema for Predictive and Generative Personalization

2026-10-01

AI Summary

This study addresses the challenge of unifying and reusing multi-source heterogeneous user histories by proposing the PerTIDE architecture. By designing an action-item shared event schema, it constructs a reusable user memory update mechanism. Integrating action gating, multi-timescale state space trajectories, and command-conditioned readout techniques, the framework enables frozen transfer of core modules across data sources and achieves unified encoding for both predictive and generative modalities, supported by theoretical invariance guarantees. Experimental results demonstrate that PerTIDE improves MRR by 15.23 points on the PENS dataset and outperforms baselines by 4.12 points on MIND, effectively validating the advantages of cross-source training.

Abstract

A user's movie, news, and dialogue histories differ in their native actions and outputs, yet each interaction supplies evidence that can update user memory. We study whether these histories can train one reusable update mechanism. An action-on-item schema pairs a mapped interaction role with a content embedding, allowing shared update parameters to operate on separate user states. We establish invariance to native relabeling, bounded state changes under item-embedding perturbations, and a pooled-training bound under explicit compatibility conditions. The Multi-Timescale State Hypothesis (MTSH) specifies how this evidence enters, persists, and is consumed; PerTIDE implements it with action gating, three state-space traces, fusion, and command-conditioned readout. On PENS, the same history encoder supports both next-news prediction and personalized headline generation. In a controlled PENS-to-MovieLens experiment, a frozen source-trained core exceeds an identically structured random core by 15.23 MRR points after fitting the same target consumer. On MIND, PerTIDE retains a 4.12-point MRR advantage over a same-input three-branch state-space control. Action, readout, and trace interventions identify complementary contributions to these gains. Together, the theory and experiments support learning history updates across compatible sources and reusing them through predictive and generative consumers.

Problem

• personalization
• cross-domain transfer
• user modeling
• predictive and generative tasks
• shared event schema

Innovation

• Action-On-Item Schema
• Shared Update Mechanism
• Multi-Timescale State Hypothesis
• PerTIDE
• Cross-domain Transfer