The thesis presents a recommender system, which is implemented using graph databases. The recommender system aims to predict the "rating" which the user would give to the element or predict which elements would be of interest to the user. There are several algorithms for recommendations. The more important approaches are: collaborative filtering, content-based filtering, and hybrid recommender systems. Graph databases are particularly suitable for such systems due to their data model. The most prominent representative is Neo4j. Based on the Neo4j system, we developed a recommender system to recommend movies (based on GroupLens data) and supported it with a web application. We used collaborative and content-based filtering. The results of the application were compared with the results of the Surprise tool. We found out that the values of MAE and RMSE are similar if we use the same algorithm.
|
