Reproduce “MMRM: A Multiplex Multimodal Representation Model for Product Ranking in E-commerce Search” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhen-Lin Chen, Maosen Sheng, Peng Lin, Jianmin Chen, Zhuojian Xiao, Dongyue Wang, Xiwei Zhao
Researchers at JD.COM developed the Multiplex Multimodal Representation Model (MMRM), a framework that enhances e-commerce product ranking by training a Multimodal Large Language Model to learn disentangled, task-specific item representations from diverse user interaction signals. The model improved predictions for user clicks, add-to-carts, and conversions in an online e-commerce search environment.
Key insights
- Diverse and heterogeneous collaborative signals contain distinct user intents that can be leveraged to learn more comprehensive and task-specific item representations.
- Disentangled, task-specific multimodal item embeddings are crucial for effectively capturing "multiplex user intents" and preventing embedding entanglement in multitask ranking.
- A unified MLLM framework can efficiently generate multiple distinct item embeddings in a single inference pass by using task-specific tokens within a shared backbone.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.