Author(s): Takahiro Komamizu
Abstract: Entity documentation is a fundamental research for utilizing Linked Data datasets. Existing entity documentation methods are based on heuristics which determine fields (predicates and/or related entities) providing contents for documents of entities. This paper releases entity documentation from heuristic methods by applying a graph proximity measurement, namely random walk with restart (RWR). This paper proposes RWRDoc, an RWR-based documentation method which generates documents of entities by weighted summation of minimal vector representations of entities. RWRDoc is a simple and general parameter-free entity documentation, and is beneficial for various applications upon entity documentations. Comprehensive experiments which apply RWRDoc for diverse applications (such as ad-hoc entity search, entity type inference, and recommender system using
Linked Data) reveal the effectiveness of RWRDoc.
Keywords: Entity documentation; Random walk with restart; Linked Data