Sign-aware recommendation based on mixed-path aggregation

doi:10.1016/j.patcog.2026.113125

Sign-aware recommendation based on mixed-path aggregation

2026-01-20

0

PRE

AI

DOI：10.1016/j.patcog.2026.113125

Original

Original request for help

Save

Abstract

En 中文

In recent years, the sign-aware graph neural network (GNN) recommender system has attracted more and more attention from researchers. Compared with unsigned GNN, sign-aware systems more accurately capture user preferences by modeling positive and negative feedback in a differentiated manner. However, there are two main challenges in existing research: First, the existing methods for modeling the propagation of negative feedback information often rely on the assumptions of social balance theory, but these assumptions may no longer hold in the context of recommender systems. Second, when modeling negative feedback information, existing methods often rely solely on the heterogeneous propagation of negative edges, ignoring the potential auxiliary information in the positive graph. To this end, this paper proposes a Recommendation Algorithm Based on Signed Mixed-Path Aggregation (Rec-SMPA). Rec-SMPA uses classic graph contrastive learning to model user positive preferences. For negative feedback modeling, this paper proposes a set of balance theories that are more suitable for the recommender system scenario and, based on it, designs a signed mixed-path aggregation module SMPA. SMPA incorporates high-order homophilic paths from positive edges to guide the propagation of negative feedback through the positive graph. Experiments on real datasets demonstrate the effectiveness of Rec-SMPA. Our implementation of Rec-SMPA is publicly available at: https://github.com/ChaoqunLiGroup/Rec-SMPA .

Sign-aware recommendation based on mixed-path aggregation

Abstract

Journal

Organization