Publication Type

Conference Proceeding Article

Version

publishedVersion

Publication Date

8-2026

Abstract

Multi-behavior recommendation improves target-behavior predic-tion by exploiting heterogeneous auxiliary feedback (e.g., view,collect, and cart), yet its robustness is often undermined by behavior-dependent noise and inconsistency. We argue that the key bottle-neck is not merely noisy behaviors, but a representation-level failurecaused by two coupled heterogeneities. First, intra-behavior rep-resentation entanglement arises when multi-hop propagationblends incidental signals with true preferences in the embeddingspace. This entanglement renders coarse spatial denoising inef-fective, since it cannot suppress noise without sacrificing weak-but-informative niche signals. Second, inter-behavior reliabilityheterogeneity complicates cross-behavior fusion, as the predic-tive value of auxiliary behaviors varies substantially across usersand contexts. Without reliability calibration, aggregation can bedominated by frequent yet untrustworthy signals, leading to target-intent drift. Existing methods typically address these issues in isola-tion and often fail when entanglement makes reliability estimationitself unstable.To resolve this robustness bottleneck, we propose DynamicSpectral Denoising with Global-Context Attention for Multi-Behav-ior Recommendation (SpectraMB), a target-oriented model that per-forms representation purification before reliability-aware fusion. Tomitigate intra-behavior entanglement, SpectraMB introduces Dy-namic Feature-Level Spectral Filtering, which re-parameterizesembeddings along the feature dimension into a feature-frequencyspace and learns view-adaptive spectral modulation end-to-endunder target supervision, enabling component-wise purificationwithout hand-crafted frequency assumptions. Built on purified rep-resentations, SpectraMB further proposes Global-Context At-tention Fusion, which uses the purified global representation asa stable context anchor to assess view compatibility and perform reliability-aware aggregation, while a residual global backbonepreserves stable collaborative structure. Extensive experiments onthree real-world datasets show that SpectraMB achieves the bestresults in most evaluation settings and exhibits improved robust-ness under noisy interactions. Our implementation is available athttps://github.com/miaomiao-cai2/SpectraMB-KDD2026.

Keywords

Multi-Behavior Recommendation, Representation Robustness, Spectral Filtering, Reliability-Aware Fusion

Discipline

Artificial Intelligence and Robotics | Databases and Information Systems

Research Areas

Intelligent Systems and Optimization

Areas of Excellence

Digital transformation

Publication

KDD '26: Proceedings of the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2, Jeju Island, South Korea, August 9-13

First Page

198

Last Page

209

Identifier

10.1145/3770855.3818191

Publisher

ACM

City or Country

New York

Additional URL

https://doi.org/10.1145/3770855.3818191

Download

Share

COinS