Gradient descent technology for sparse vector learning in ontology algorithms

Abstract

In the information retrieval field, ontology plays a crucial role in searching for information that is highly semantic similar to the query concept in origin. It also plays a major roles in returning the results to the user. The main purpose of ontology algorithm lies in obtaining the function for calculating the similarity between concepts. One learning technology in applying ontology is to obtain a ontology function f:V Æ R In this structure, it maps each vertex to a real number and the similarity between two concepts is then determined by the difference of their real numbers. In this article, we report a sparse vector learning algorithm for ontology similarity measure and ontology mapping in terms of gradient descent and iterative computation. The ontology penalty term in optimization model is reformulated by virtue of dual norm, and then the smooth approximation to this penalty term is obtained. The main procedure of our algorithm is iterative computation based on gradient descent tricks. Results achieved in the simulation experiments show that the new proposed algorithm is highly efficient and accuracy in ontology similarity measure and ontology mapping in multiple disciplines.

Keywords:

• Ontology
• similarity measure
• ontology mapping
• sparse vector
• gradient descent