Beyond Word Clouds: Exploring Your Car’s Problems in 3D

When your car makes a strange rattling sound, you might head online to see if others have had the same issue. But if you’ve ever waded through thousands of consumer complaint reports, you know the frustration. You’re often met with long, dry lists of text or simple “word clouds” that lack context.

While word clouds show you what words are popular, they don’t tell you where those problems actually happen on the vehicle. A new approach called Descriptive Non-photorealistic Rendering (d-NPR) is changing that. By mapping thousands of real-world text reports directly onto a 3D model, researcher Meng-Wei Chang and Supervisor Dr. Christopher Collins have created a way to “see” data in its physical home, with a focus on helping consumers make better, more informed choices when they purchase a vehicle

Bringing Text to Life

The core idea behind d-NPR is that many documents describe physical objects. If you are reading about a “broken handle” on a “toaster,” your brain naturally visualizes where that handle is.

The system uses a smart hierarchy to link words to 3D parts. For example, it knows that a “bonnet” or “cowlings” is a synonym for a “hood” and that both are parts of a “car”. By scanning over 600,000 reports from the National Highway Traffic Safety Administration (NHTSA), the system counts how often parts are mentioned and highlights them on a 3D car model.

Seeing the “Pop-Out” Effect

To make the data easy to read at a glance, the system uses a “yellow-to-red” colour scale. The more complaints a part has, the redder and more solid it becomes. Parts with no issues are shown as simple outlines so they don’t distract you, but still provide context.

This allows for a “pop-out” effect: when you look at the model, your eyes are immediately drawn to the most significant failures.

The d-NPR interface: Text data is transformed into a 3D visualization where problematic parts “pop out” based on how often they are mentioned in reports.

Tools for Digital Mechanics

The research introduced several interactive widgets to help users dig deeper into the data:

• The Interactive Lens: Much like a digital magnifying glass, you can move a lens over the car to see specific details about the parts underneath it.
• Heatmaps: These small grids show how complaints change over time. It’s easy to see if a part fails more in the spring or if a problem is getting worse year by year.
• Comparison Mode: This is perhaps the most useful tool for car buyers. You can compare two vehicles — like a Honda Civic and a Toyota Corolla — and the system will highlight which has a higher failure rate for specific parts with green and pink outlines.

Comparison mode: The coloured outlines indicate which of the two vehicles has more reported issues for specific components.

Putting it to the Test: What Users Thought

12 participants were recruited to perform real-world tasks, such as choosing the more reliable car between two options. Participants loved the car model’s “relatable” nature. One user noted that it was much “nicer to look at a picture than a bunch of numbers.” While some found the heatmaps a bit detailed, the ability to “dissect” the car with the lens was a huge hit.

A participant using the lens to examine the vehicle.

Real-World Insights: The Toyota Recall

To test the system, researchers looked at the famous Toyota recall of 2009–2010. While the engine usually dominates car complaints, during this period, two other parts clearly “popped out” on the 3D model: the brakes and the accelerator.

Get vialab research’s stories in your inbox

Join Medium for free to get updates from this writer.

Remember me for faster sign in

The first thing we noticed is that the engine, which usually had one of the highest occurring components in the complaint reports, no longer dominated the visualization, but the brakes and accelerator pop out instead.

A closer examination with the lens widget raised more questions, and the heatmaps showed that there are two outlier months in which a large number of complaints were registered. We found a massive “spike” in complaints during February and March 2010, suggesting a trigger that caused a loss of consumer confidence, which in turn led to an over-reporting of problems. This is supported by the sharp drop-off after March 2010.

Future Directions

While the current prototype focuses on cars, the researchers believe this “spatial” approach to text can be applied to almost anything.

• Building Maintenance: Visualizing repair logs on a 3D model of an office building or a landmark like the Sydney Opera House.
• Consumer Products: Seeing exactly where users are reporting cracks or failures on a new smartphone or home appliance.
• Sentiment Analysis: Future versions could use advanced AI to distinguish between people complaining about a part and people praising it, adding another layer of colour to the 3D view.

By moving away from boring lists and into 3D space, d-NPR makes complex data relatable, helping everyday people make more informed decisions, whether they’re buying a car or just trying to figure out what that “clunking” sound is.

This post is based on the paper “Exploring Entities in Text with Descriptive Non-photorealistic Rendering” by Meng-Wei Chang and Christopher Collins, published in IEEE Transactions on Visualization and Computer Graphics (2013).