A taxonomy based semantic similarity of documents using the cosine measure

Abstract

In this paper, we present a new method for calculating semantic similarities between documents. This method is based on cosine similarity calculation between concept vectors of documents obtained from a taxonomy of words that captures IS-A relations. The calculation of semantic similarities between documents is a very time consuming task, since it is necessary first to calculate semantic similarities between each pair of words that appear on different documents. In this paper, we present a new method to calculate semantic similarities between documents which results in faster computational time. Both a taxonomy based semantic similarity and cosine similarity are employed. First, the concept vectors of documents are obtained by extending the terms in the document vectors with their corresponding IS-A concepts. Cosine similarity is then calculated between those concept vectors of documents. Thus, the overall similarity between documents is a combination of cosine similarity and semantic similarity. The proposed semantic similarity is tested in document clustering problem. The experimental results show that our method achieves a good performance.