News clustering using linked data resources and their relationships

Text clustering or document clustering is the task of identifying and grouping text documents that are about the same topic. This is especially important for the ever growing Web where the number of free-text documents just keep increasing. News clustering is a special task in this domain in which the goal is to classify news documents by topic. Earlier solutions on this problem utilized ``bag of words'' approach in which documents are represented with words and their frequencies in documents, and the clustering task measures the similarity of documents using this representation. However, this approach does not take into consideration the meaning or the importance of words and ambiguity in words is not resolved. We present a new approach to document or news clustering, we utilize ``linked data''. We map words or phrases in news documents to their real-world counterparts in ``linked data'' knowledge bases such as DBpedia and represent documents with linked data entities they have. Then we cluster documents using these entities and their category hierarchy similarities. Evaluation results show that our approach performs better than the bag of words approach.