Recommender Systems has become more and more sophisticated in their theoretical approaches and implementations over the last decade, while the scope of their application has extended at the same time. This makes it an interesting research field in the world of Data Mining.
This thesis gives a summary about the most common types of Recommender Systems, about how and based on what we group them into different classes, and what are the main advantages and disadvantages of them. It also investigates several implementation and evaluation techniques of Recommender Systems. Solutions using a provided framework and others built on Neo4j’s graph database are also introduced thoroughly from building a graph database model to implementing filtering methods using the power of the graph data structure. The characteristics and functionalities of Neo4j’s property graph database are part of the thesis too, as well as the angles of designing a graph model based on the features of Neo4j. Recommender algorithms using different content based filtering and collaborative filtering approaches are both explained and tested from theoretical and implementation perspective.
Finally, in pursuit of improving the developed standalone baseline methods, two additional hybrid solutions with detailed comparison to the baseline methods are also included in this thesis. Moreover, among other things, the summary closing this thesis contains the results of the hybrid Recommender System and further improvement possibilities of the proposed algorithms.