Search papers, labs, and topics across Lattice.
This paper introduces a dual-enhancement method for product bundling that combines interactive graph learning with LLM-based semantic understanding to address cold-start item recommendations. A Dynamic Concept Binding Mechanism (DCBM) translates graph structures into natural language prompts, aligning domain-specific entities with LLM tokenization. Experiments on three benchmarks show 6.3%-26.5% improvements over state-of-the-art baselines, demonstrating the effectiveness of bridging interactive graphs and LLMs.
LLMs can now effectively recommend product bundles, even for cold-start items, by translating interactive graphs into natural language prompts.
Product bundling boosts e-commerce revenue by recommending complementary item combinations. However, existing methods face two critical challenges: (1) collaborative filtering approaches struggle with cold-start items owing to dependency on historical interactions, and (2) LLMs lack inherent capability to model interactive graph directly. To bridge this gap, we propose a dual-enhancement method that integrates interactive graph learning and LLM-based semantic understanding for product bundling. Our method introduces a graph-to-text paradigm, which leverages a Dynamic Concept Binding Mechanism (DCBM) to translate graph structures into natural language prompts. The DCBM plays a critical role in aligning domain-specific entities with LLM tokenization, enabling effective comprehension of combinatorial constraints. Experiments on three benchmarks (POG, POG_dense, Steam) demonstrate 6.3%-26.5% improvements over state-of-the-art baselines.
