Abstrakt
Recently, named entity recognition tools tend to disambiguate recognized named entities on a very detailed level. Instead of elementary types (e.g.
Person or Location), they assign concrete identifiers, trying to distinguish even different entities having same name and type (e.g. cities with the same name in different countries). We introduce a novel method for this kind of named entity disambiguation exploiting structural dependencies of recognized entities.
We analyse the co-occurrence of disambiguated entities in the backing knowledge base and use this information to improve results of existing named entity disambiguation approaches. A model for co-occurrence representation is proposed and evaluated based on a dataset that we mine from Wikipedia.