TR-11-10: Measuring Semantic Similarity using a Multi-Tree Model

Abstract

Recommender systems and search engines are examples of systems that have used techniques such as Pearson’s product-momentum correlation coefficient or Cosine similarity for measuring semantic similarity between two entities. These methods relinquish semantic relations between pairs of features in the vector representation of an entity. This paper describes a new technique for calculating semantic similarity between two entities. The proposed method is based upon structured knowledge extracted from an ontology or a taxonomy. A multi-tree concept is defined and a technique described that uses a multi-tree similarity algorithm to measure similarity of two multi-trees constructed from taxonomic relations among entities in an ontology. Unlike conventional linear methods for calculating similarity based on commonality of attributes of two entities, this method is a non-linear technique for measuring similarity based on hierarchical relations which exist between attributes of entities in an ontology. The utility of the proposed model is evaluated by using Wikipedia as a collaborative source of knowledge.

TR-11-10.pdf