Semantic Classification

A key challenge in the semantic web is the mapping between different concepts. Many techniques for such mapping exist, but most of them induce a one-to-one mapping, which does not seem to correspond to real world problems. This project proposes a new approach, which tries to use the power of machine learning, and in particular classification algorithms, to solve the mapping task. It introduces a new semantic similarity metric which is used with semantic metadata and classification algorithms. The approach is tested in a real world dataset. Pre-processing of the dataset took place, and in particular feature selection, extraction and representation was implemented, for both content-based and semantic features. The documents of the dataset were classified using the content-based features, the semantic ones, and their combination. The results were compared and they gave as an insight of how semantic features can affect classifiers and traditional features.