Measuring Semantic Similarities to Classify Instances for Ontology Learning

Ontology learning refers to generating scalable ontologies based on the web of documents. It includes two main processes: first, extracting concepts and their semantic relationships; second, classifying instances based on the extracted concepts. In this paper, we proposed an effective approach called Max Similarity Min Distance Algorithm (MSMDA) to address the second process. Traditionally, the instances are classified by a classifier, which is trained by a group of samples. The training process is time-consuming and labor-intensive. To avoid the training process, our approach uses Normalized Compression Distance (NCD) to measure the similarities to classify instances. We improved the traditional NCD algorithm by calculating the semantic similarities between instances. We conducted the empirical experiments based on two datasets. The experiment results indicate that MSMDA outperforms the traditional NCD algorithm in terms of