Due to the needs to improve the information search process, new strategies have been created to enhance searches. The semantic search performs the search by means of meaning instead of literals. The semantic search in unstructured documents requires to formalize knowledge through an annotation semantic process. Some annotation proposals use natural language processing tools, ontologies to link document terms; others use the similarity of entities through the weight of the edges, association between pair of concepts or the ontology structure. In this paper we present an alternative for semantic annotation in unstructured documents by semantic context extraction of entities. In the approach we detect the named entities through a data dictionary created from Wikipedia and link the instances in the ontology. The context extraction strategy is based on the concepts similarity; each term is associated with an instance of the ontology and the similarity between relationships explicit is measured by the combination of two types of measures: the association between each pair of concepts and the weight of the relationships. The approach was tested with two ontologies and two datasets in news and business, respectively.