SynGR: Unleashing the Potential of Cross-Modal Synergy for Generative Recommendation

Generative Recommendation (GR) has emerged as a promising paradigm by formulating item recommendation as a sequence-to-sequence generation task over item identifiers. Recent studies have incorporated multimodal signals to provide richer token-level evidence for generation. However, existing approaches largely rely on alignment-centric fusion and underexplore synergistic information across modalities. In practice, synergistic information plays a critical role in capturing emergent item properties that cannot be inferred from any single modality alone. Such properties encode intrinsic item semantics and guide user preferences, enabling models to move beyond surface-level feature matching. To address this limitation, we propose \textbf{SynGR}, a synergistic generative recommendation framework that explicitly encourages the exploitation of cross-modal dependencies during generation. By constraining overreliance on dominant modalities, SynGR enables the model to capture emergent item semantics beyond shared or modality-specific signals. Extensive experiments across three benchmark datasets demonstrate that SynGR achieves superior performance.

Many recommendation systems use product images and text descriptions to predict what a user may like next. However, these systems often treat the two types of information mainly as matching signals, instead of understanding what new meaning appears when they are combined. For example, a product photo may show stylish materials and design, while its text may mention the brand and price; together, they can reveal a sense of luxury or quality that neither source fully explains alone. In this work, we study how to help generative recommendation models better use this combined image-text meaning. We propose SynGR, a framework that reduces the model’s tendency to rely too much on one dominant source of information and encourages it to consider how different modalities complement each other. This matters because users often make decisions based on subtle combinations of visual style, product details, and personal taste. Experiments on three real-world recommendation datasets show that SynGR gives more accurate recommendations than strong existing methods. Our results suggest that future recommender systems can become more reliable by learning not only from images or text separately, but from the richer meaning created when they work together.