Déjà Vu of Strange Stickers! Enhancing Out-of-Distribution Robustness in Sticker Retrieval via Cross-Modal Intent Alignment

ACL ARR 2025 February Submission5456 Authors

Abstract: The rapid evolution of digital communication has amplified the demand for sticker retrieval systems that can match vivid stickers as carriers to satisfy the user's expressive needs. However, real-world sticker retrieval faces significant out-of-distribution (OOD) challenges from unseen queries and stickers, due to the diverse user expression habits and sticker visual representations. The OOD issues often result in the retrieval of irrelevant or inappropriate stickers, negatively impacting the user experience. Inspired by symbolic interactionism in cognition, this paper proposes XAlign-SR to improve OOD robustness in sticker retrieval by aligning abstract expressive intent between queries and stickers across different modalities. We construct two OOD sticker retrieval benchmarks that simulate real-world OOD queries and sticker scenarios. Both online and offline experiments demonstrate that our approach significantly outperforms prevailing baselines.

Paper Type: Long

Research Area: Multimodality and Language Grounding to Vision, Robotics and Beyond

Research Area Keywords: Image text matching, Cross-modal application, Cross-modal information extraction

Contribution Types: NLP engineering experiment, Data resources

Languages Studied: English, Chinese

Submission Number: 5456