Prototype-Aligned Federated Soft-Prompts for Continual Web Personalization

arXiv cs.LG / 3/31/2026

📰 NewsSignals & Early TrendsIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper proposes ProtoFed-SP, a privacy-conscious, parameter-efficient prompting framework for continual web personalization under non-stationary user behavior and privacy constraints.
ProtoFed-SP uses a dual-timescale design: a fast, sparse short-term soft prompt for session intent and a slow long-term soft prompt anchored to a server-side prototype library.
Long-term prototypes are updated via differentially private federated aggregation, and user queries are routed to the top relevant prototypes to compose personalized prompts on the fly.
Experiments on eight benchmarks show improved ranking and engagement metrics (e.g., NDCG@10 +2.9% and HR@10 +2.0% over the strongest baselines) alongside reduced forgetting while maintaining accuracy within practical DP budgets.
The authors frame the approach as a controllable way to balance stability (retaining long-term preferences) and plasticity (adapting to new intents) using a transparent prompting interface anchored to semantic prototypes.

Abstract

Continual web personalization is essential for engagement, yet real-world non-stationarity and privacy constraints make it hard to adapt quickly without forgetting long-term preferences. We target this gap by seeking a privacy-conscious, parameter-efficient interface that controls stability-plasticity at the user/session level while tying user memory to a shared semantic prior. We propose ProtoFed-SP, a prompt-based framework that injects dual-timescale soft prompts into a frozen backbone: a fast, sparse short-term prompt tracks session intent, while a slow long-term prompt is anchored to a small server-side prototype library that is continually refreshed via differentially private federated aggregation. Queries are routed to Top-M prototypes to compose a personalized prompt. Across eight benchmarks, ProtoFed-SP improves NDCG@10 by +2.9% and HR@10 by +2.0% over the strongest baselines, with notable gains on Amazon-Books (+5.0% NDCG vs. INFER), H&M (+2.5% vs. Dual-LoRA), and Taobao (+2.2% vs. FedRAP). It also lowers forgetting (AF) and Steps-to-95% and preserves accuracy under practical DP budgets. Our contribution is a unifying, privacy-aware prompting interface with prototype anchoring that delivers robust continual personalization and offers a transparent, controllable mechanism to balance stability and plasticity in deployment.