Increasing performance in Force Directed Graphs

: an application in cancer support service provision profiling

Abstract

This thesis aims to show the progress made in the development and testing of a 3D data visualisation tool using Force Node Graphs (FNG). This project was run with stakeholders from the Digital Health and Innovation Centre (DHI) and Macmillan Cancer Support (Macmillan). It uses the FNG to display association rule minded data from an eHNA data set provided by Macmillan. The objectives were to build a visualisation tool able to depict these data in the form of an interactive graph. Work with stakeholders to design and implement key interactions to filter and navigate the data and to determine if the visualisation tool is of value to and usable by Macmillan workers.

To have an interactive, real time rendering of these data, technical improvements were necessary to improve the frame rate and the CPU usage of the program. This took the form of the Barnes-Hut algorithm which was used to approximate the Node-to-Node physics calculations in the graph as well as a conversion to using a GPU Compute Shader to improve the speed of the Edge Physics Calculations. A range of options to filter the data were added to the tool though co design with the stakeholders. To test for value for Macmillan a series of online user workshops were presented that allowed three groups of Macmillan staff from different departments to view use cases and comment on the tool. They were asked to take a System Usability Scale (SUS) test after the workshop. Afterwards a thematic analysis of the workshop recordings was done to tease more insight to the thoughts on the tool.

The results of the technical improvements were mixed. The Barnes-Hut implementation was hugely successful at improving performance with an increase from 5 FPS to 40FPS on average. The GPU Compute shader was less successful, being slightly slower than the CPU. This is hypothesised because of the scale of the dataset, where a bigger dataset would have yielded better results. The user testing results were also mixed. The SUS scores only showed a quarter of participants thought the tool was successful, but the answers for each question showed promise. The thematic analysis showed participants found the tool features to be well integrated, but concerns arose over the relevance of the data being displayed. A gap was shown between the more and less technically proficient jobs within Macmillan with different responses to the ease-of-use questions. Further workshops would lead to improvements in the feature set and help bridge the gap between the different sets of workers.

Date of Award	27 Oct 2023
Original language	English
Awarding Institution	Abertay University
Supervisor	Ruth Falconer (Supervisor), James Bown (Supervisor) & Kean Lee Kang (Supervisor)