Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Personalized Prompt for Sequential Recommendation

Yiqing Wu, Ruobing Xie, Yongchun Zhu, Fuzhen Zhuang, Xu Zhang, Leyu Lin, Qing He

TL;DR

This work tackles cold-start challenges in sequential recommendation by introducing Personalized Prompt-based Recommendation (PPR), which generates user-profile–driven soft prefix prompts and injects them into pre-trained sequential models. A prompt-oriented contrastive learning regime augments training to robustly align prompt-augmented representations under data sparsity. The approach demonstrates strong improvements in few-shot and zero-shot settings, across multiple backbone models (e.g., SASRec, CL4SRec) and tasks (including cross-domain recommendation and user profile prediction). Notably, PPR offers parameter-efficient tuning via two modes (PPR(light) and PPR(full)) and shows universality and robustness across datasets. The results suggest a practical, scalable path to leverage large pre-trained recommender models through personalized prompts in real-world cold-start scenarios.

Abstract

Pre-training models have shown their power in sequential recommendation. Recently, prompt has been widely explored and verified for tuning in NLP pre-training, which could help to more effectively and efficiently extract useful knowledge from pre-training models for downstream tasks, especially in cold-start scenarios. However, it is challenging to bring prompt-tuning from NLP to recommendation, since the tokens in recommendation (i.e., items) do not have explicit explainable semantics, and the sequence modeling should be personalized. In this work, we first introduces prompt to recommendation and propose a novel Personalized prompt-based recommendation (PPR) framework for cold-start recommendation. Specifically, we build the personalized soft prefix prompt via a prompt generator based on user profiles and enable a sufficient training of prompts via a prompt-oriented contrastive learning with both prompt- and behavior-based augmentations. We conduct extensive evaluations on various tasks. In both few-shot and zero-shot recommendation, PPR models achieve significant improvements over baselines on various metrics in three large-scale open datasets. We also conduct ablation tests and sparsity analysis for a better understanding of PPR. Moreover, We further verify PPR's universality on different pre-training models, and conduct explorations on PPR's other promising downstream tasks including cross-domain recommendation and user profile prediction.

Personalized Prompt for Sequential Recommendation

TL;DR

This work tackles cold-start challenges in sequential recommendation by introducing Personalized Prompt-based Recommendation (PPR), which generates user-profile–driven soft prefix prompts and injects them into pre-trained sequential models. A prompt-oriented contrastive learning regime augments training to robustly align prompt-augmented representations under data sparsity. The approach demonstrates strong improvements in few-shot and zero-shot settings, across multiple backbone models (e.g., SASRec, CL4SRec) and tasks (including cross-domain recommendation and user profile prediction). Notably, PPR offers parameter-efficient tuning via two modes (PPR(light) and PPR(full)) and shows universality and robustness across datasets. The results suggest a practical, scalable path to leverage large pre-trained recommender models through personalized prompts in real-world cold-start scenarios.

Abstract

Pre-training models have shown their power in sequential recommendation. Recently, prompt has been widely explored and verified for tuning in NLP pre-training, which could help to more effectively and efficiently extract useful knowledge from pre-training models for downstream tasks, especially in cold-start scenarios. However, it is challenging to bring prompt-tuning from NLP to recommendation, since the tokens in recommendation (i.e., items) do not have explicit explainable semantics, and the sequence modeling should be personalized. In this work, we first introduces prompt to recommendation and propose a novel Personalized prompt-based recommendation (PPR) framework for cold-start recommendation. Specifically, we build the personalized soft prefix prompt via a prompt generator based on user profiles and enable a sufficient training of prompts via a prompt-oriented contrastive learning with both prompt- and behavior-based augmentations. We conduct extensive evaluations on various tasks. In both few-shot and zero-shot recommendation, PPR models achieve significant improvements over baselines on various metrics in three large-scale open datasets. We also conduct ablation tests and sparsity analysis for a better understanding of PPR. Moreover, We further verify PPR's universality on different pre-training models, and conduct explorations on PPR's other promising downstream tasks including cross-domain recommendation and user profile prediction.
Paper Structure (30 sections, 7 equations, 5 figures, 8 tables)

This paper contains 30 sections, 7 equations, 5 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Methodology
  4. Preliminaries
  5. Background of Prompt in NLP
  6. Goals of PPR
  7. Notions of PPR
  8. Overall Framework
  9. Warm-training Stage of PPR
  10. Personalized Prompt-tuning
  11. Personalized Prompt Generator
  12. Prompt-tuning of PPR
  13. Prompt-oriented Contrastive Learning
  14. Prompt-based Augmentation
  15. Behavior-based Augmentation
  16. ...and 15 more sections

Figures (5)

  • Figure 1: An example of prompt-tuning in recommendation.
  • Figure 2: Overall architecture of (a) the conventional fine-tuning, and (b) our Personalized prompt-based recommendation.
  • Figure 3: Ablation study of PPR(light) and PPR(full) on (a) CIKM, and (b) AliEC.All improvements are significant over baselines (t-test with p ${<}$ 0.05).
  • Figure 4: model performs under different noise on AliEC dataset
  • Figure 5: Relative improvements of PPR models over fine-tuning with different degrees of data sparsity in CIKM dataset.All improvements are significant over baselines (t-test with p ${<}$ 0.05).