Finding Good URLs: Aligning Entities in Knowledge Bases with Public Web Document Representations

In this paper we address the novel task of mapping entities from a knowledge base to public web documents. This task is of relevance for aligning structured data with web documents, e.g., for the purpose of providing equivalent human readable representations of entities or to detect and propagate changes on the web to the knowledge base. An alternative interpretation of the task is to find good public URLs for the entities in a knowledge base. In order to address the task, we adapt and investigate several approaches based on web search and link network analysis. We compare nine approaches including ordinary web search for the text label of an entity as well as link analysis strategies like HITS authority ranking or PageRank. We evaluate the approaches under the aspect of identifying URLs of documents which are good representations of a given entity. In general, our experiments show a significant advantage of label based web search over all other methods. Furthermore, we introduce a filtering technique leveraging semantic typings to boost the performance of virtually all methods.