MOON3.0: Reasoning-aware Multimodal Representation Learning for E-commerce Product Understanding

Abstract

With the rapid growth of e-commerce, exploring general representations rather than task-specific ones has attracted increasing attention. Although recent multimodal large language models (MLLMs) have driven significant progress in product understanding, they are typically employed as feature extractors that implicitly encode product information into global embeddings, thereby limiting their ability to capture fine-grained attributes. Therefore, we argue that leveraging the reasoning capabilities of MLLMs to explicitly model fine-grained product attributes holds significant potential. Nevertheless, achieving this goal remains non-trivial due to several key challenges: (i) long-context reasoning tends to dilute the model's attention to salient information in the raw input; (ii) supervised fine-tuning (SFT) primarily encourages rigid imitation, limiting the exploration of effective reasoning strategies; and (iii) fine-grained details are progressively attenuated during forward propagation. To address these issues, we propose MOON3.0, the first reasoning-aware MLLM-based model for product representation learning. Our method (1) employs a multi-head modality fusion module to adaptively integrate raw signals; (2) incorporates a joint contrastive and reinforcement learning framework to autonomously explore more effective reasoning strategies; and (3) introduces a fine-grained residual enhancement module to progressively preserve local details throughout the network. Additionally, we release a large-scale multimodal e-commerce benchmark MBE3.0. Experimentally, our model demonstrates state-of-the-art zero-shot performance across various downstream tasks on both our benchmark and public datasets.

Figures (6)

• Figure 1: Overall results on all the downstream tasks.
• Figure 2: Comparison of fine-grained attributes among the query, the ground-truth product (denoted as "GT Product"), and a similar product. "$\to$" denotes the cosine similarity between their representations.
• Figure 3: Pipeline of our MOON3.0. Each element in the training triplet (query, positive, and negative items) is represented in three modalities: image-only, text-only, and multimodal (image and text). "Attr." denotes product attributes.
• Figure 4: Multi-head Modality Fusion module.
• Figure 5: A training sample of our MBE3.0 benchmark. Each training sample includes the image and title of the query, positive item, and negative item, as well as three types of attributes: image-derived attributes (“Img Attr.”), text-derived attributes (“Text Attr.”), and multimodal attributes (“MM Attr.”, the union of image and text attributes).