Behavior-Contextualized Item Preference Modeling for Multi-Behavior Recommendation

TL;DR

BCIPM addresses noise from auxiliary behaviors in multi-behavior recommendation by learning item-aware preferences within each behavior and aggregating only target-relevant signals for recommendation, while using auxiliary behaviors to train embeddings and network parameters. It introduces a three-part architecture consisting of Embedding Pre-training on a unified multi-behavior graph, the Behavior-Contextualized Item Preference Network (BIPN) for behavior-contextualized item preferences, and a GCN Enhancement Module to reinforce target-behavior representations. Experiments on four real-world datasets demonstrate state-of-the-art performance and robust noise reduction from auxiliary behaviors, with comprehensive ablations validating each module. The work advances practical multi-behavior recommendation by exploiting item-aware preferences and behavior-specific filtering, particularly benefiting sparse-target scenarios.

Abstract

In recommender systems, multi-behavior methods have demonstrated their effectiveness in mitigating issues like data sparsity, a common challenge in traditional single-behavior recommendation approaches. These methods typically infer user preferences from various auxiliary behaviors and apply them to the target behavior for recommendations. However, this direct transfer can introduce noise to the target behavior in recommendation, due to variations in user attention across different behaviors. To address this issue, this paper introduces a novel approach, Behavior-Contextualized Item Preference Modeling (BCIPM), for multi-behavior recommendation. Our proposed Behavior-Contextualized Item Preference Network discerns and learns users' specific item preferences within each behavior. It then considers only those preferences relevant to the target behavior for final recommendations, significantly reducing noise from auxiliary behaviors. These auxiliary behaviors are utilized solely for training the network parameters, thereby refining the learning process without compromising the accuracy of the target behavior recommendations. To further enhance the effectiveness of BCIPM, we adopt a strategy of pre-training the initial embeddings. This step is crucial for enriching the item-aware preferences, particularly in scenarios where data related to the target behavior is sparse. Comprehensive experiments conducted on four real-world datasets demonstrate BCIPM's superior performance compared to several leading state-of-the-art models, validating the robustness and efficiency of our proposed approach.

Figures (6)

• Figure 1: Examples of user behavior influenced by their decision factors.
• Figure 2: Overview of our proposed BCIPM.
• Figure 3: Structure of the Behavior-Contextualized Item Preference Network.
• Figure 4: Behavior-contextualized item preference network analysis ("r.al." indicates remove both pre-filtering layer and post-filtering layer, "r.pr." means remove pre-filtering layer, "r.po." means remove post-filtering layer, and "our" is our method).
• Figure 5: Effectiveness analysis of auxiliary behavior ("b.r." indicates simultaneously eliminate auxiliary behavior from both two modules, "n.r." means exclude auxiliary behaviors solely from the BIPN module, and "all" implies keep them all).