Motivation

In recent years, the rapid advancement of multimodal large models and the increasing demand for sophisticated information retrieval systems have highlighted the importance of integrating diverse modalities into real-world applications. Multimodal Information Retrieval (MIR) aims to address challenges where textual, visual, and other content modalities coexist, pushing the boundaries of search, recommendation, and information processing capabilities.

To address these challenges, we have organized two major themes, each consisting of two tracks:

1. Multimodal Document Retrieval Challenge Track:
   • MMDocIR: Multi-Modal Retrieval for Long Documents.
   • M2KR: Open-Domain Vision Retrieval Benchmark.
2. Multimodal CTR Prediction Challenge Track:
   • Multimodal Item Embedding.
   • Multimodal CTR Prediction.

These themes aim to foster innovation in:

• Developing retrieval systems capable of handling diverse multimodal data.
• Exploring efficient representation learning methods tailored for multimodal content.
• Encouraging solutions that bridge academic research and industrial practices in multimodal retrieval and recommendation.

Organizers

Xuri Ge
Shandong University

Junchen Fu
University of Glasgow

Jingbiao Mei
University of Cambridge

Fajie Yuan
Westlake University

Jieming Zhu
Huawei Noah's Ark Lab

Jiabing Yan
Huawei

Kuicai Dong
Huawei Noah's Ark Lab

Zhaocheng Du
Huawei Noah's Ark Lab

Yong Liu
Huawei Noah's Ark Lab