RELIN: Relatedness and Informativeness-based Centrality for Entity Summarization

Linked Data is developing towards a large, global repository for structured, interlinked descriptions of real-world entities. An emerging problem in many Web applications making use of data like Linked Data is how a lengthy description can be tailored to the task of quickly identifying the underlying entity. As a solution to this novel problem of entity summarization, we propose RELIN, a variant of the random surfer model that leverages the relatedness and informativeness of description elements for ranking. We present an implementation of this conceptual model, which captures the semantics of description elements based on linguistic and information theory concepts. In experiments involving real-world data sets and users, our approach outperforms the baselines, producing summaries that better match handcrafted ones and further, shown to be useful in a concrete task.