Artificial Intelligence Knowledge Graph Data Visualisation

National Project Research in China

Platform

Unity | C#

3D.JS | Javascript

Python

Sketch

Affiliation:

Chinese Academy of Sciences and Communication University of China

Main tasks：

Interaction system design
Partial system development
Writing 6000+ reports

Background

This research was done in collaboration with the Chinese Academy of Sciences. Based on artificial intelligence data processing technology, the key ending of how users interact with the data helps users to search for data accurately and intelligently. As the knowledge graph itself is extremely professional, the AI technology system relies on real-time analysis and processing of big data, with many data dimensions, huge data magnitude, a complex mesh structure formed between knowledge nodes, and the accuracy of the knowledge nodes to be ensured, there are huge technical barriers to a truly interactive and functional mapping system, with practical problems in design and user experience such as single vision, weak readability, biased technology, low efficiency in obtaining effective information, and difficulty in interaction.

Research Goal

Provides an intuitive, visually appealing layout for data presentation and explores innovative interactive forms of mapping. Users can get a general view of the data and quickly clarify the logical relationships behind the data structure visually, making it easy for them to trace the search results back to their roots, while the good interactive format greatly enhances the user experience. The specific objectives are as follows:

Quantitative variation of knowledge: design presentation of data magnitudes and functional presentation to individual knowledge nodes.
AI Knowledge extraction: visualization of the knowledge extraction process and the extraction results.
Visual design for knowledge reasoning.

Research Workflow

Research Challenges

There are three ways to achieve sub-node dispersion by applying a force or modifying the position of the node in real-time, which have been investigated: using AddForce to achieve sub-node distribution by applying physical pressure, using Lerp functions to update the exact start and end points in real-time, and using the absolute position updated in real-time to move the sub-node to achieve dispersion.

We are instantiating child nodes: the maintenance and dynamics of the framework. We want to determine the parent and child objects when inputting data. Instantiating child objects helps the migration and optimization of the overall mapping framework.

Two ways of instantiating and implementing child node clustering and dispersion are currently being explored: generating child objects by establishing a polar coordinate system and converting the position coordinates in real-time through the tangent, and generating child objects and then determining the position of the center of the circle by calculating the angle parallels.

Research Design

The existing engine Unity3D was chosen as the development tool for the prototype development of this design solution, which can quickly build an interactive prototype and is compatible with a variety of interaction methods, which can be adapted to the mouse and keyboard and can also be applied to different interaction tools such as VR to visualize the data.

To meet the visual effect of presenting a relatively large amount of data, the space of the scene is built using the metaphorical meaning of the vast starry sky, with each element carrying specific data symbolized and built up by the many planets in the universe as the scene contains galaxies of different light levels and colors. The background effects are created using the Visual Effect Graph system. This system can be used to create large-scale visual effects for Unity projects, using the GPU to simulate particle behavior, allowing for far more particles to be manufactured than the built-in particle system.