A service-oriented framework for personalized recommender systems using a colour-impression-based image retrieval and ranking method

This paper points out that achievements in the field of multimedia analysis and retrieval represent an important opportunity for improvement of recommender system mechanisms. Online shopping systems use various recommender systems; however a study of different approaches has shown that they do not exploit the potential of information carried by multimedia product data for product recommendations. We demonstrate how this can be accomplished by a personalized recommender system framework that is based on a method of analysis of colour features of entity images. Colour-features are based on image colour histograms, psychological properties of colours and a learning mechanism. We have developed a service-oriented framework for a personalized recommender system that is based on incorporation of this method into a highly interactive business process model. The framework is designed in a generic way and can be applied to an arbitrary domain. It is based on service-oriented architecture in order to promote its flexibility and reuse, which is important when applying it to existing recommender system environments. An experimental study was performed for the domain of travel agency. The framework provides several important advantages, such as automatic creation of entity image meta-data which is based on colour-based image analysis and extraction of their semantic properties, user-interaction based learning, dynamic selection and presentation ordering of entity images, and feedback for creation of base image entity sets.