Collective Entity Linking Based on DBpedia

With the rapid development of Internet, lots of web data are published by internet users. This situation causes tremendous entities appear on the web. However, because of variety and ambiguity of natural language, one entity usually has multiple expressions. To know the actual meaning of one document, it is important to solve the problem of entity ambiguity. Entity linking is a good solution for entity disambiguation. It links one entity to one entrance of a resource to help users grasp the actual meaning of this entity. For the reason that traditional entity linking methods cannot acquire high performance in both accuracy and efficiency, we propose a novel entity linking algorithm. This algorithm is mainly divided into three steps. It first generates candidate entities for each mention in documents via heuristic-based rule. Then we leverage the relationship between entities in the knowledge base and use them to construct a semantic entity graph to connect all the related candidate entities. Finally we give a score to measure the possibility of one entity to be an entrance for one mention and choose the one with the highest score as the best assignment. Experimental results show that our entity linking algorithm performs well in both accuracy and efficiency.