Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Generative Framework for Personalized Sticker Retrieval

Changjiang Zhou, Ruqing Zhang, Jiafeng Guo, Yu-An Liu, Fan Zhang, Ganyuan Luo, Xueqi Cheng

TL;DR

PEARL introduces a generative, personalized sticker retrieval framework that encodes user-group information and decodes property identifiers with intent-aware guidance. By learning discriminative user-group embeddings through three prediction tasks and leveraging CoT-driven intent ranking for property decoding, PEARL achieves substantial offline and online gains over state-of-the-art baselines. The approach demonstrates strong practical value for chat-based sticker discovery, reducing retrieval errors while improving click-through and user satisfaction, albeit with considerations around privacy, modality, and LLM costs. Overall, PEARL advances personalized generative retrieval in a multimodal, real-world setting, offering a scalable pathway for intent-aligned, user-aware content retrieval.

Abstract

Formulating information retrieval as a variant of generative modeling, specifically using autoregressive models to generate relevant identifiers for a given query, has recently attracted considerable attention. However, its application to personalized sticker retrieval remains largely unexplored and presents unique challenges: existing relevance-based generative retrieval methods typically lack personalization, leading to a mismatch between diverse user expectations and the retrieved results. To address this gap, we propose PEARL, a novel generative framework for personalized sticker retrieval, and make two key contributions: (i) To encode user-specific sticker preferences, we design a representation learning model to learn discriminative user representations. It is trained on three prediction tasks that leverage personal information and click history; and (ii) To generate stickers aligned with a user's query intent, we propose a novel intent-aware learning objective that prioritizes stickers associated with higher-ranked intents. Empirical results from both offline evaluations and online tests demonstrate that PEARL significantly outperforms state-of-the-art methods.

A Generative Framework for Personalized Sticker Retrieval

TL;DR

PEARL introduces a generative, personalized sticker retrieval framework that encodes user-group information and decodes property identifiers with intent-aware guidance. By learning discriminative user-group embeddings through three prediction tasks and leveraging CoT-driven intent ranking for property decoding, PEARL achieves substantial offline and online gains over state-of-the-art baselines. The approach demonstrates strong practical value for chat-based sticker discovery, reducing retrieval errors while improving click-through and user satisfaction, albeit with considerations around privacy, modality, and LLM costs. Overall, PEARL advances personalized generative retrieval in a multimodal, real-world setting, offering a scalable pathway for intent-aligned, user-aware content retrieval.

Abstract

Formulating information retrieval as a variant of generative modeling, specifically using autoregressive models to generate relevant identifiers for a given query, has recently attracted considerable attention. However, its application to personalized sticker retrieval remains largely unexplored and presents unique challenges: existing relevance-based generative retrieval methods typically lack personalization, leading to a mismatch between diverse user expectations and the retrieved results. To address this gap, we propose PEARL, a novel generative framework for personalized sticker retrieval, and make two key contributions: (i) To encode user-specific sticker preferences, we design a representation learning model to learn discriminative user representations. It is trained on three prediction tasks that leverage personal information and click history; and (ii) To generate stickers aligned with a user's query intent, we propose a novel intent-aware learning objective that prioritizes stickers associated with higher-ranked intents. Empirical results from both offline evaluations and online tests demonstrate that PEARL significantly outperforms state-of-the-art methods.

Paper Structure

This paper contains 23 sections, 23 equations, 6 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Problem Statement
  3. Methodology
  4. Overview
  5. Model Architecture
  6. Personalized representation learning
  7. Intent-aware model training
  8. Experimental Settings
  9. Experimental Results
  10. Main results
  11. Ablation studies
  12. Efficiency analysis
  13. Online tests
  14. Related work
  15. Conclusion
  16. ...and 8 more sections

Figures (6)

  • Figure 1: The overview of PEARL.
  • Figure 2: The learning of user-specific representation.
  • Figure 3: Case study on retrieved results of online system and PEARL.
  • Figure 4: Examples for distinct properties of stickers in the corpus.
  • Figure 5: Case study for the user query "Angry" by male users aged 0-19.
  • ...and 1 more figures