Representing Textual Passages as Graphs to Support Question Answering

Abstract

As technology and its capabilities are profoundly increasing, the means of communication between humans has become immensely easier. Technological advances are ever more applicable with Artificial Intelligence and the processing of natural language data. However, a problem that is worth pursuing is the concept that machines processing natural language data have little comprehension during question answering. Machines often have difficulty understanding information because data are usually not represented in a comprehensible manner. To take a step toward solving this problem, this thesis will explore a new, automated way to represent any short news passage in the form of a graph. Such graphs are useful because they represent the most amount of information while being compact and leading to accurate, efficient answers. The ability to see relationships throughout entire passages, having properties that make question answering possible, and being able to graph any short news passage bring immense value to this project. This project carries significance because of the fact that it is interfaceable with other systems, simplifying work by serving as a driver and being able to be combined with additional tools. The Natural Language Tool Kit, Neo4j Database Software, Stanford Core Natural Language Processing, and the Python programming language are all tools that were used in completing this project.