BRIDGE: Bundle Recommendation via Instruction-Driven Generation

TL;DR

Bundle recommendation is challenged by sparse user-bundle interactions and the need to model cohesive item sets. BRIDGE introduces an instruction-driven, distant-supervision-inspired framework that generates pseudo-ideal bundles from correlation-based item clusters and user history, then retrieves real bundles via a Transformer-based generator and a dual-metric ranking strategy. The model jointly optimizes clustering, generation, and ranking losses ($L = L_G + L_C + L_R + \\lambda \\|\\theta\\|_2^2$) to produce accurate top-$K$ recommendations, and it shows consistent, statistically significant improvements over state-of-the-art baselines on five public datasets. This approach broadens the search space beyond predefined bundles, bridging user imagination and available bundles and improving practical bundle recommendation performance.

Abstract

Bundle recommendation aims to suggest a set of interconnected items to users. However, diverse interaction types and sparse interaction matrices often pose challenges for previous approaches in accurately predicting user-bundle adoptions. Inspired by the distant supervision strategy and generative paradigm, we propose BRIDGE, a novel framework for bundle recommendation. It consists of two main components namely the correlation-based item clustering and the pseudo bundle generation modules. Inspired by the distant supervision approach, the former is to generate more auxiliary information, e.g., instructive item clusters, for training without using external data. This information is subsequently aggregated with collaborative signals from user historical interactions to create pseudo `ideal' bundles. This capability allows BRIDGE to explore all aspects of bundles, rather than being limited to existing real-world bundles. It effectively bridging the gap between user imagination and predefined bundles, hence improving the bundle recommendation performance. Experimental results validate the superiority of our models over state-of-the-art ranking-based methods across five benchmark datasets.

Figures (7)

• Figure 1: An example of instruction-driven bundle recommendation
• Figure 2: The end-to-end architecture of BRIDGE
• Figure 3: Impact of $\#$Encoder/Decoder block.
• Figure 4: Impact of max-context length $T$.
• Figure 5: Impact of trade-off coefficient $\alpha$.