I will illustrate this blog post using a well-known dataset of Game of Thrones character interactions: In this blog post, I will explain how I built a graph visualization tool and some of the challenges I had to overcome.
DIRECTED GRAPH BUILDER SOFTWARE
That includes machine learning, data visualization, third-party software integration, etc. I work on the R&D team at Dataiku and we are always looking to provide users with novel ways to analyze their data within our data science platform. When looking at a graph of social media accounts and their interactions, for example, seeing many accounts pointing towards the same account could potentially mean that they are fake. Graph visualization has plenty of useful applications, such as detecting fake social media accounts. Displaying a graph as nodes and edges is quite intuitive and allows us to easily spot patterns. I wanted to find an easy way to visualize a table as a graph network to get insights about complex connected data - insights that are not easy to get when only looking at the raw dataset, such as finding clusters and outliers.
![directed graph builder directed graph builder](https://flutterawesome.com/content/images/2020/08/TopDownTree.png)
The underlying structure of our table now becomes a graph: a set of nodes (e.g., persons) that are connected to each other through edges (e.g., relationships). However, these rows can be related (e.g., some people may have relationships with others). In data science, most data is represented in the form of tables that consist of columns and rows and we consider rows as independent (e.g., each row corresponds to a single person).