A citation recommendation model employing knowledge graph embedding

Relevant Work Discovery Difficulty: The growth and topical diversity of scientific publications makes it difficult for researchers to locate truly relevant works.

Underused Heterogeneous Citation Network Structure: Existing citation recommendation models under-exploit the rich multi-relational structure of heterogeneous citation networks.

Weak Semantic Interaction Modeling: Existing models inadequately model semantic interactions among diverse entities.

Limited Recommendation Explainability: Existing citation recommendation models lack mechanisms to identify influential factors or provide transparent explanations.

Relevant Work Discovery Difficulty: The growth and topical diversity of scientific publications makes it difficult for researchers to locate truly relevant works.

CR-KGEB Citation Recommendation Framework: CR-KGEB is an encoder–decoder citation recommendation framework integrating RotatE knowledge-graph embeddings, SPECTER content representations, and a BiLSTM attention module over a k-partite graph.

Joint Signal Modeling with Attention: The method jointly models an author’s published articles, citation history, and candidate papers while dynamically weighting salient signals.

CR-KGEB Citation Recommendation Framework: CR-KGEB is an encoder–decoder citation recommendation framework integrating RotatE knowledge-graph embeddings, SPECTER content representations, and a BiLSTM attention module over a k-partite graph.

Joint Signal Modeling with Attention: The method jointly models an author’s published articles, citation history, and candidate papers while dynamically weighting salient signals.

Improved DBLP-V12 Recommendation Accuracy:

Improved DBLP-V13 Recommendation Accuracy:

Interpretable Recommendations: