HGIB improves multi-behavior recommendation by adding information-bottleneck preservation and compression losses plus a learnable graph refinement encoder to hierarchical models, with reported gains on academic and industrial datasets.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.IR 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Hierarchical Graph Information Bottleneck for Multi-Behavior Recommendation
HGIB improves multi-behavior recommendation by adding information-bottleneck preservation and compression losses plus a learnable graph refinement encoder to hierarchical models, with reported gains on academic and industrial datasets.