Document clustering and topic discovery based on semantic similarity in scientific literature

Unlabeled document collections are becoming increasingly common and mining such databases becomes a major challenge. It is a major issue to retrieve relevant documents from the larger document collection. By clustering the text documents, the documents sharing similar topics are grouped together. Incorporating semantic features will improve the accuracy of document clustering methods. In order to determine at a sight whether the content of a cluster are of user interest or not, topic discovery methods are required to tag each clusters identifying distinct and representative topic of each cluster. Most of the existing topic discovery methods often assign labels to clusters based on the terms that the clustered documents contain. In this paper a modified semantic-based model is proposed where related terms are extracted as concepts for concept-based document clustering by bisecting k-means algorithm and topic detection method for discovering meaningful labels for the document clusters based on semantic similarity by Testor theory. The proposed method is compared to the Topic Detection by Clustering Keywords method using F-measure and purity as evaluation metrics. Experimental results prove that the proposed semantic-based model outperforms the existing work.