Year: 2024
Type: conference-paper
Abstract: In the realm of knowledge graph completion, multi-relational link prediction presents a formidable challenge due to the intricate interactions and implicit information shared between entities. Many existing methodologies predominantly build upon binary relationship models, often overlooking the distinct roles that entities assume in multi-relational contexts and their subsequent interactions. To bridge this gap, we introduce a novel approach termed Role Semantic-Aware multi-relational link prediction (RSP). This model emphasizes the roles that entities assume within relationships, leveraging key-value pairs of entity roles to encapsulate the implicit information and profound semantics inherent in multi-relational data. By decomposing intricate relationships into multiple sub-relationships, RSP adeptly captures nuanced interactions between entities. Furthermore, it employs projection matrices to amalgamate role and entity embedding vectors, thereby effectively discerning the complex links between entities. Empirical evaluations on two publicly accessible multi-relational datasets reveal that RSP surpasses other benchmark models, particularly demonstrating superior performance when handling multi-relational data with heightened density.
Authors: Xiaohui Dong, Xingyu Qi, Chenmeng Ma, Xiaochao Dang
Institution: Northwest Normal University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Award: G8154757074
OpenAlex ID: W4409095071