Using AI tools with concept mapping activities and scoring
Embedded Files
by Elise Barrella, Cheryl Bodnar & Maria-Isabel Carnasciali
by Elise Barrella, Cheryl Bodnar & Maria-Isabel Carnasciali
February 2025
During the development of this toolkit, which began in 2021, large language models (LLMs) like ChatGPT or CoPilot have been introduced and rapidly advancing in popularity and capability. So we want to share examples of how GenAI may make it easier to create, collect, or score concept maps. We'll also point out some of the questions and challenges related to using AI for grading or scoring concept maps.
Examples include:
Using AI to create and iterate a concept map.
Using AI to "read" and give feedback on handwritten or digital concept maps.
Using AI to score using one of the validated methods.
And more as AI applications continue to evolve!
Drawing concept maps with AI - still a work-in-progress
Drawing concept maps with AI - still a work-in-progress
What if you ask ChatGPT to draw you a concept map about Heat Transfer? How does it do?
Well, it might first try to tell you how to make one yourself.
Then, it might be able to create a list...
Maybe you're interested in "separation processes".
You said:
Can you generate a concept map on separation processes?
ChatGPT said:
Certainly! Here's a concept map for Separation Processes in Chemical Engineering. It focuses on key principles, methods, and applications
[…]
Would you like me to create a visual representation of this concept map?
You said:
Can you make this an image
And it created what might be called a mind-map. Sort of?
So you try to revise.
You said:
Could you use words on the linking lines and remove pictures and just use words?
Traditional scoring with AI
Traditional scoring with AI
As a test, we tried scoring a concept map with the traditional method using Google's Gemini. We used the example below, which you may remember from the second module on Assigning concept maps.
KEY_fill-in-map_Mod2.pdfInitially, we started with this prompt:
I've copied in a concept-map. I'd like you to:
1) count the number of concepts (A concept is a word or short phrase contained in a bubble. Do not count the main topic)
2) count the number hierarchies (Each branch from the main topic represents a different “hierarchy” of related terms. Each hierarchy is composed of one or more propositions.) Also determine the length of the highest hierarchy (length means the number of levels in a hierarchy)
3) count the number of cross-links. (A cross-link connects two separate hierarchies and shows a relationship between different conceptual threads.)
Generate a score where there's 1 point for each concept, 5 for the highest hierarchy, and 10 for each cross-link.
Then, generate some formative feedback for what this map does well and what it might have missed
The initial score calculation was not correct:
Concepts: 30 concepts * 1 point/concept = 30 points
Highest Hierarchy: 1 hierarchy * 5 points/hierarchy = 5 points
Cross-links: 2 cross-links * 10 points/cross-link = 20 points
Total Score: 30 + 5 + 20 = 55 points
So we went through a few iterations to correct the number of concepts (actually 22), the highest hierarchy (3), and the number of cross-links (0). You can see an example of how the number of cross-links was corrected.
Try the exercise for yourself using your AI tool of choice. Experiment with multiple concept maps and compare your hand scoring (or our semi-automated scoring tool) to the AI. Maybe test concept maps that were drawn by hand instead of with the CmapToolsTM software, like our example.
How'd it do?
While GenAI is evolving rapidly, there are still some things, like drawing a concept map, that needs a little work. A specialized software tool like CmapToolsTM or templates in Miro are still probably better options for creating a concept map that has complexity and connections. However, students may have discovered that they can use AI to get started with a list of concepts. You can always have them draw by hand...
Using AI to "read" and give feedback on student concept maps
Using AI to "read" and give feedback on student concept maps
Screenshot of example output from the Concept Map Analyzer (credit: Doug Melton)
GenAIs like ChatGPT are getting better and better at "reading" and extracting information from images.
This means, whether your students are hand drawing or using a digital tool to create concept maps, a GPT may be able to help you scan concept maps and draw observations about how your students perceive EM.
With training, an AI tool can even offer students feedback or a qualitative "score". It can also help students reflect on how to improve their concept map and thus their ideas about the target topic.
Dr. Doug Melton, KEEN Program Director, has created a "Concept Map Analyzer" using ChatGPT that is able to extract topics from an uploaded concept map image and provide an evaluation of how comprehensive and connected the representation is (see the example evaluation and feedback). It is also programmed to provide feedback on ways to improve the concept map and discussion prompts to help students reflect on how the topic relates to the 3C's of Curiosity, Connections, and Creating Value.
Learn more about how the Concept Map Analyzer works and try it for yourself!
Note: Doug provides the disclaimer that the analyzer uses a GPT and not a validated scoring method like the ones we've shared in this toolkit and research articles. The real value is in helping students reflect on their concept map and continuing making connections.
Where do we go from here?
We want to learn with you and share successful practices for using AI to assist with scoring of concept maps in particular. If you'd like to share a case study, please contact us or connect with us on the EngineeringUnleashed platform.
At the 2025 KEEN National Conference, we engaged faculty in thinking about how AI tools might open up new options for creating and scoring concept maps. Check out the conversation: Link to KNC card for AI & Concept Mapping Workshop
Page updated
Report abuse