Turning Noise into Value: Uncovering Service Preferences from Ambiguous Interaction in E-commerce

TL;DR

NoVa reframes multi-behavior e-commerce service recommendation as a positive-unlabeled problem, addressing the misinterpretation of auxiliary interactions as negatives. It introduces an adversarial distribution alignment module to discover high-confidence false negatives and a semantic-guided bias correction to suppress noise, integrating these signals into a PU-aware ranking objective. Extensive experiments on three real-world datasets demonstrate consistent improvements over state-of-the-art baselines, and ablations confirm the complementary roles of alignment and filtering. The approach offers a scalable, robust way to leverage ambiguous user behavior to better capture latent service preferences in sparse environments.

Abstract

In e-commerce service recommendation, utilizing auxiliary behaviors to alleviate data sparsity often relies on the flawed assumption that auxiliary behaviors that fail to trigger target actions are negative samples. This approach is fundamentally flawed as it ignores false negatives where users actually harbor latent intent or interest but have not yet converted due to external factors. Consequently, existing methods suffer from sample selection bias and a severe distribution shift between the auxiliary and target behaviors, leading to the erroneous suppression of potential user needs. To address these challenges, we propose a Noise-to-Value Adapter (NoVa), an e-commerce service recommendation framework that re-examines the problem through the lens of positive-unlabeled learning. Instead of treating ambiguous auxiliary behaviors as definite negatives, NoVa aims to uncover high-quality preferences from noise via two key mechanisms. First, to bridge the distribution gap, we employ adversarial feature alignment. This module aligns the auxiliary behavior distribution with the target space to identify high-confidence false negatives, which are instances that statistically resemble confirmed target behaviors and thus represent latent conversion intents. Second, to mitigate label noise caused by accidental clicks or random browsing, we introduce a semantic consistency constraint. This mechanism implements semantic-aware filtering based on the content similarity of services, acting as a bias correction step to filter out low-confidence interactions that lack semantic relevance to historical user preferences. Extensive experiments on three real-world datasets demonstrate that NoVa outperforms state-of-the-art baselines.

Figures (5)

• Figure 1: An illustrative example of the positive-unlabeled nature in service interactions, distinguishing high-confidence false negatives from label noise.
• Figure 2: The overall framework of the proposed NoVa model. Grounded in a positive-unlabeled learning perspective, NoVa consists of two core modules: (1) adversarial feature alignment for bridging distribution gaps and recovering high-confidence false negatives; and (2) semantic-aware filtering for suppressing label noise via consistency constraints. The refined value signals extracted by these modules are then integrated into the target service recommendation task.
• Figure 3: Ablation studies of NoVa, where IJCAI stands for IJCAI-Contest, and Retail stands for Retail Rocket.
• Figure 4: Hyperparameter $\mu$ and $\beta$ analysis of NoVa.
• Figure 5: Visualization of sample distributions in the embedding space. The yellow points represent the latent positive set recovered by NoVa, illustrating how they bridge the distribution gap between observed positives and negatives.