Proximity points approach for building recommendation system for the “Open Karelia” information system

Abstract

This paper describes proximity points approach for building recommendation systems. This approach is applicable for information systems which content (objects) is a semi-structured data with weak links. The described approach was used for implementing the "Open Karelia" recommendation system. The paper gives details about the proximity criterion selection and testing for specific domain field and use case. The work also enlightens of the concrete comparing criterion choosing approach and gives implementation details about using Sphinx search engine in case of full text fields.