URL: http://www.ijicic.org/ijicic-11-07081.pdf ------------------------------------ Resumen: In recent times there has been an interest in developing automatic techniques to create and/or populate knowledge bases (KB) and ontologies. One initiative along this line is the Knowledge Base Population (K BP) track of the Text Analysis Conference (TA C), which offers a framework for the evaluation of automatic systems designed to populate ontologies using information found in unstructured text. In K BP, the knowledge base population process has been divided into two complementary tasks: entity linking, whose goal is to detect mentions in text to instances in a reference KB, and slot filling, which extracts from text facts about the previously detected instances and adds these facts to the KB. In this paper, an automatic, unsupervised algorithm for entity linking is presented. The main novelty of the proposed approach, named WikiIdRank, comes from applying a Page Rank-like algorithm to a network of instance co-occurrences to address the task. The algorithm was implemented and evaluated in the K BP 2010 with positive results. Furthermore, an analysis of the impact of instance co-occurrence information in the entity linking process was carried out, and indicates a gain of accuracy around 16% compared with a baseline approach relying on information retrieval techniques.