arXiv：cs.LG（机器学习，全量分类）· Parthiv Chatterjee, Kashish Kanjaria, Vashisth Purani, Sourish Dasgupta, Tanmoy Chakraborty·· 14 小时前AI 评分38
PerTIDE：面向预测与生成式个性化的统一事件模式与偏好流
Action-On-Item Preference Flow: A Shared Event Schema for Predictive and Generative Personalization
AI 导读
研究者提出 Action-On-Item 偏好流与 Multi-Timescale State Hypothesis（MTSH），并实现 PerTIDE，用动作门控、三条状态空间轨迹、融合与命令条件读出，让同一历史编码器同时支持下一新闻预测和个性化标题生成。
正文
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.
| Comments: | Accepted to NeurIPS 2026. Author-prepared archival version with expanded discussion and interpretation |
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.01375 [cs.LG] |
| (or arXiv:2610.01375v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.01375 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Sourish Dasgupta [view email]
[v1]
Thu, 1 Oct 2026 09:42:46 UTC (1,668 KB)
来源：arXiv：cs.LG（机器学习，全量分类） · arxiv.org