Clustering Graphs -- Applying a Label Propagation Algorithm to Detect Communities in Graph Databases

TL;DR

This paper presents a full-stack system that applies a label propagation algorithm to detect communities in graph databases, specifically using academic publication data to identify clusters of related research topics.

Contribution

It introduces a practical implementation of community detection in graph databases with a web interface, integrating data manipulation, API, visualization, and clustering algorithms.

Findings

Effective clustering of academic publications into research communities

Successful integration of graph database, API, and visualization tools

Demonstrated utility of label propagation for community detection in real-world data

Abstract

In the last few decades, Database Management Systems (DBMSs) became powerful tools for storing large amount of data and executing complex queries over them. In the recent years, the growing amount of unstructured or semi-structured data has seen a shift from representing data in the relational model towards alternative data models. Graph Databases and Graph Database Management Systems (GDBMSs) have seen an increase in use due to their ability to manage highly-interconnected, continuously evolving data. This thesis is a documentation of the work done in implementing a system to identify clusters in graph modeled data using a Label Propagation Community Detection Algorithm. The graph was built using datasets of academic publications in the field of Computer Science obtained from dblp.org . The system developed is a FullStack WebApp consisting of a web-based user interface, an API and the data (nodes, edges, graph) stored in a Graph Database Management System (GDBMS). Described in this document are: - the process of manipulation pre-import and import of the data in a Graph Database Management System (GDBMS) such as ArangoDB, creation of nodes, relations (edges) between the nodes and a graph composed of these nodes and edges; - the GraphQL API implemented in NodeJS to request data from the Graph Database Management System (GDBMS); - the frontend interface made with TypeScript and React consisting of the search functionalities and ability to visualize results in Cytoscape Network Graphs; - the Label Propagation Community Detection Algorithm execution on the graph, the found clusters which are stored and visualized to the user whenever requested. This thesis hopes to contribute with a practical hands-on approach on the graph representation, integration and analysis of interconnected data.