Classifying Wikipedia entities into fine-grained classes

Recognition of named entities (people, companies, locations, etc) is an essential task of text analytics. We address the subproblem of this task, namely, named entity classification. We propose a novel approach that constructs an effective fine-grained named entity classifier. Its key highlights are semi-automatic training set construction from Wikipedia articles and additional feature selection. We justify our solution by creating 18-class classifier and demonstrating its effectiveness and efficiency.