Graph-based recommendation system design with comparative similarity analysis

Recommendation systems are designed to enhance user experiences in digital environments by suggesting the most relevant products to users. These systems first emerged in the 1990s, and with the increasing use of online platforms, research in this field has significantly grown. A graph database is a NoSQL (Not only SQL) database frequently used for storing relational data, as it demonstrates better performance compared to other types of databases in handling large volumes of relational data. In this thesis, a comparative similarity analysis was conducted using a recommendation system designed on a graph database. The model was built on the Neo4j graph database using the MovieLens dataset, where user-based and item-based filtering were applied. In the model, Cosine, Euclidean, Manhattan, Chebyshev, and Jaccard similarity measures were calculated separately for both users and movies, and these similarity measures were utilized for rating prediction. Score predictions were made using the weighted sum method and were subsequently compared with the actual scores. The error values were measured using the metrics of Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE), Normalized Mean Absolute Error (NMAE), and Root Mean Square Error (RMSE). Based on these error metrics, the most suitable similarity method was identified.