Introducing Our AI Model to understand Lived Experiences
Our Patent-Pending AI Model to Classify Feedback
Urban planners and designers: how do you design solutions for communities?
Ideally, the process looks simple as this:
Research: Develop a plan for your research study
User Feedback: Create a community engagement plan to get community feedback
Analysis: Analyze data from community feedback
Prototype: Develop feasibility studies and reports highlighting areas of improvement for public projects.
But here’s the problem - turning tons of qualitative data (from interviews, meetings, and surveys) into actionable, useful next steps is not easy. After talking to over 100 researchers, we learned that translating feedback into solutions is the most challenging part of the process, no question. Sentiment analysis is helpful for designers and planners, but knowing negative, positive, and neutral does not help you know what to do next.
The Old Way
Today’s process for analyzing community feedback is tedious, at best. If a design or planning firm is lucky, they have an analyst or engineer who has the capacity to build a limited model for topic labeling. Most teams at firms are actually taking survey responses and transcriptions (from recorded meetings) and classifying individual responses by hand, using software like Excel or Google Sheets. Maybe, even going old-school and using Post-its. After getting past the information paralysis, analysis takes a lot of focus, time, and energy.
This can take weeks, but we knew there was a way to cut it down to minutes. The answer lies in text classifiers.
Enter: Text Classifiers
Think about the spam folder in your email inbox. It automatically hides the seemingly endless and very dangerous emails you receive (free car giveaways, hot singles in your area, your long-lost wealthy auntie, you know how it is). But, how does it work?
Most spam emails are sorted by an automated text classifier. Learning from the billions of spam emails sent daily (yes, billions) the text classifier figures out which keywords and phrases are signs of spam. Then, it analyzes each email you receive to sort them into the folder they belong in. This is more efficient than you doing it by yourself, and it’s accurate.
Using the same method, we realized that text classifiers can help urban designers find solutions to community problems in the community feedback itself. When analyzing data, designers want to identify the community paint points and gains, which text classifiers can do automatically. But, somebody just needed to build one.
So we built one.
Say Hello to The Dubois™
Introducing our patent-pending text classifier: The Dubois™. Our team developed this AI model using Natural Language Processing (NLP). As a text classifier, The Dubois™ can take the text or voice data you collected and sort conversational community feedback into the following categories:
| Pain Points | Gains |
|---|---|
| Are Experienced as:: Behaviors (actions, patterns, reactions in community) Affect (responses to infrastructure or policy) Conditions (physical infrastructure of community) |
Are Identified as: Solutions (community-requested outcomes or ideas) |
It’s like magic! But really, it’s just an algorithm built by some very smart people.
How The Dubois™ Works
After importing data into co:census, The Dubois™ starts categorizing community feedback in minutes. Once the data is sorted, sentiment analysis and trends become clear. Users can then filter the feedback for “Solutions” to only see the feedback that includes ideas from community members. This is vital to design using the voice of your community. Teams can also filter through Insights to see “Behaviors”, a powerful tool to understand traffic patterns that are observed by actual residents.
The combined power of these tools bridge the gap between communities and decision makers. It gives urban designers and planners more time back in their days to learn from their communities. It lets them spend more time reaching the communities that are historically excluded from community input. It lets them make a bigger impact on the world.
At co:census, we focus on putting tools into the hands of intentional urban designers to help them create real reflections of the community they’re designing. The Dubois™ is an important piece of the puzzle, enabling users to make better and more equitable decisions from community feedback, quickly.
We’ve spent nearly three years building The Dubois™ into the powerful tool it is today. Roughly 100 hours (and counting!) have gone into the design, data labeling, and training of the algorithm. Every working hour has resulted in our current 76% accuracy rate. As we continue to train The Dubois™ , its accuracy will only improve.
Now, when urban designers are looking for real community insight on a new project, they can immediately see the proposed solutions that used to be buried in the data.
It’s the magnet that finds the needle in the haystack.