Document classification utilising ontologies and relations between documents

Two major types of relational information can be utilized in automatic document classification as background information: relations between terms, such as ontologies, and relations between documents, such as web links or citations in articles. We introduce a model where a traditional bag-of-words type classifier is gradually extended to utilize both of these information types. The experiments with data from the Finnish National Archive show that classification accuracy improves from 70% to 74% when the General Finnish Ontology YSO is used as background information, without using relations between documents.