.png){kind=icon}
.png){kind=icon}
Return
PAGSL: Path-Augmenting Graph Structure Learning for explainable scholar recommendation
.png){kind=icon}
PRE
AI
DOI：10.1016/j.ipm.2026.104826.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• We propose a novel explainable scholar recommendation framework, PAGSL, which unifies graph structure refinement with interpretable semantic path modeling, effectively addressing both noise reduction and user- centric explanation in academic social networks. • We develop a path-aware relation learning module that integrates augmented path trees with heterogeneous encoding. This heterogeneous encoding architecture preserves the branching propagation of academic influence, overcoming the expressive bottlenecks of sequential paths in capturing complex contextual nuances. • We design a topological path explanation strategy based on hierarchical decomposition guided mask learning and path score optimization. Our module enables the generation of explanatory paths characterized by high fidelity, thereby achieving dual improvement in both recommendation precision and explainability. • Through experimental evaluations on three real-world academic social networks, the results reveal that PAGSL surpasses existing state-of-the-art methods, effectively bridging the gap between scholar recommendation effectiveness and explanatory quality.
Keywords:
explainable scholar recommendation
graph structure learning
path-aware relation learning
topological path explanation
heterogeneous encoding
Journal
I
IF:
6.9
Papers:
325
Citations:
0