I’m in the process of pulling together a presentation for next month’s Data Visualization Summit in Boston, a conference organized by the Innovation Enterprise team. The event attracts 150-200 industry folks to see what can be done using data visualization approaches. I committed to share some insights on using network visualization to visually analyze customer behavior, and after a few weeks of tossing ideas around, have settled on a final approach. Now it’s time to actually put some data together and create some impressive visualizations for the presentation.
The end goal is to share how interactive network graphs can be used to tap into customer insights from several angles. There are three levels of analysis I’m hoping to share with the group, using some wholly fictitious data for a consumer products company. In order, the three stages are:
- Create a network that displays customer purchase patterns by product, providing a quick yet insightful visual overview showing who buys what, and how different products intersect with one another. For example, we might see a strong visual correlation where the shoppers who purchase Product A also buy Product D, but rarely purchase Product C. This in itself should provide some value, although other visualization methods could also perform this task, albeit in a less elegant fashion.
- Stage two is to focus on overall customer satisfaction levels (with the company rather than individual products), and potentially on an individual product basis, although this gets a bit more complex to execute. Through the effective use of color, we can scale satisfaction levels using the original purchase graph, thus providing a more powerful visual image. Decision makers can now easily view multiple attributes in a single visualization, something that is often difficult to achieve using conventional charts or tables.
- The third stage providers viewers with the ability to see actual customer comments, including summarized versions of said comments. This will enable analysts and decision makers to discover common themes that may be linked to low (or high) satisfaction levels. Again, this would be a challenging task using other visualization approaches, but can be handled effectively using well designed network graphs.
So how do we pack all this information into a single, easy to use visualization? For starters, we employ Gephi, the powerful network graph tool that allows us to convert purchase behavior data into nodes and edges that define our network graph. We can use Gephi to define the best layout for our dataset, create specific groups, make adjustments to sizes and colors, and so on. From there, we’ll be exporting the graph file using the Gexf-JS Web Viewer plugin, which will enable user interactivity through a browser. Finally, we can tweak some of the settings to deliver an attractive, intuitive, highly useful network graph visualization.
Before I forget, I must mention that the brilliant Aylien text analysis service will be used to analyze and categorize our customer comments. The results can then be included in our Gephi source files, adding another layer or two of rich insights to the data and ultimately the network graph. Integrating text analysis results with transactional customer information is an area that continues to evolve, and is a key component in understanding the present and predicting the future of customer behavior.
I hope to share the final deck at a future point, or at least the network graph that makes up the primary component of the presentation. Until then, happy visualizing!