Files
Abstract
RegPattern2Vec is a novel algorithm used for knowledge graph embedding. It effectively samples a large Knowledge Graph to learn node embeddings, while capturing the semantic relationships between the nodes of graph. This thesis proposes an implementation of the RegPattern2Vec algorithm as a custom plugin in the Neo4j graph database. This plugin generates random walks, which can then be used to generate vector embeddings. The plugin accepts a regular expression transformed Finite Automata as one of the inputs, and generates random walks based on this Finite Automata as output, all of which is implemented in Java. The generated random walks through this procedure effectively capture the semantic relationship between the graph nodes with minimum prior knowledge and human involvement. The plugin is then successfully tested on a subset citation network dataset and Olympics Results dataset and the results are documented.