Art recommendation APP

Abstract

A good recommendation system is one that provides accurate and useful recommendations to users. User clustering is employed by numerous systems to provide recommendations based on user similarity. The system in this thesis will implement an item clustering technique so as to tackle issues related to inaccurate outcomes as a result of user clustering. It will utilize a hybrid of collaborative and content-based recommendation approaches by employing such clustering and rulemining techniques. An element of two interesting concepts, namely, serendipity and competition also form part of this thesis to give the prototype an added edge. Personal recommendations and a point awarding system combined in a mobile app bring together a challenging project and new interesting research values. Results from both accuracy and user-satisfaction tests indicate the system is returning fitting results. Initial recommendations were based on users with similar interests. An improvement of 6% user satisfaction was achieved when building user profiles using item clustering, rule- mining, serendipity and a random recommendation. The average F-measure was found to be 0.79. Final results indicate that using such methods, the aim and objectives set prior to the development can be achieved.