Many recommendation systems are built after the release of the software, on the already used technologies. These systems’ performance may not be as satisfying as needed. They are not as fast as necessary, and their maintainability can be difficult, since they were not built for these kind of problems originally.
In this document I examine, how to build recommendation engines on a graph data structure, and on a relational data structure, and compare their differences, and their performance. The recommendation algorithms that I use are the most common algo-rithms, that are used in graph based recommendation systems, such as Person similarity or Cosine Distance calculations. These algorithms are implemented on the relational en-gine too, to get similar results.
This thesis work gives a basic knowledge about how graph based recommendation sys-tems work. It shows their performance, and also states what problems can appear while building a recommendation system on older, not “job-specific” technologies.
The results show that the graph based engine is not faster than the relational one, but it might prove beneficial to use them. Namely, their usability and maintainability are much better, than those of the relational implementation.