Deriving Insights from Graph Databases using Embedding Techniques

The term Big Data is characterized by features such as Volume, Velocity, Veracity, Variety, etc. It is not possible to handle the Big Data system using only the traditional Relational Database Management System (RDBMS). NoSQL (Not Only SQL) Database Systems are generally used to handle big datasets. In the modern connected world, the relationships between entities are equally important as the entities themselves. Therefore, social networking sites like Twitter, Facebook, and LinkedIn, among others, make use of Graph Databases to model the relationships that exist between entities/users. In a Graph database, there are nodes representing the entities, edges and the relationships between them. Some of the predominant Graph Databases in use are FlockDB, Neo4j, AllegroGraph, J anusGraph, DGraph, etc. In this research work, the N eo4j Graph Database platform is chosen, and experimental evaluations are conducted to demonstrate that Graph Embedding techniques can be used to map each entity (typically nodes) into a fixed-length vector representation. The performance of the FastRP (Fast Random Projection) graph embedding algorithm is compared with that of the node2vec graph embedding algorithm. The results of the FastRP embedding algorithm are used to perform experiments that showcase how the results of Graph Embedding can be utilized as input for the Machine Learning (ML) pipeline to derive meaningful insights.