Presentation
Using LinkIt Constrained Concept Mapping to make Anesthesiology Training More Effective
SessionPoster Session 2
DescriptionTraining for a career in medicine is expensive in terms of both time and money (American Medical Association, 2023). For example, becoming a physician typically requires four years of undergraduate study, four years of medical school, and three to seven years of residency. The cost often exceeds $200,000 USD. Physician assistants (PAs) complete a bachelor’s degree and then a 2–3 year PA program, requiring about 6–7 years total, with program costs often between $70,000–$90,000.
Further, once in practice, healthcare professionals continue their education via continuous medical education (CME) credits. These requirements vary by state and certifying body, so details depend on where you practice. For example, Texas MD/DO physicians must do 48 credits every 24 months.
Medical training is also time-consuming and expensive for the instructors who deliver lessons, provide feedback and grade exams. To alleviate some of these inefficiencies, we have been developing a tool based on mental models and concept mapping.
Mental models are internal, simplified representations of a topic that people use to reason, predict, and learn. They enable individuals to simulate outcomes, evaluate alternatives, and guide decision-making. This framework, first formalized by Johnson-Laird, suggests that reasoning does not rely on formal logic but rather on constructing and manipulating these internal models to “read off” possible conclusions (Johnson-Laird, 1983; Wykes, 1984). Mental models are not static; they evolve as individuals incorporate new information, supporting both problem-solving and adaptive learning processes (Li & Maani, 2011). They also enable people to run “mental simulations” of events, which underpin complex tasks such as conditional reasoning, spatial reasoning, and decision-making in uncertain environments (Jahn, Johnson-Laird, & Knauff, 2005). Beyond supporting individual cognition, mental models are dynamic tools for learning—changing existing knowledge structures to accommodate new experiences and helping to avoid errors in judgment that stem from incomplete or flawed internal representations (Schaffernicht et al., 2021).
Mental models can be represented as networks, where nodes correspond to concepts and labeled links capture the relationships among those concepts. This graph-based perspective highlights how knowledge structures are interconnected, allowing individuals to navigate between concepts and reason about their relations. For instance, in spatial reasoning tasks, mental models have been modeled using labeled graphs in which nodes represent objects and edges encode spatial relations between them (Geibel, 1998). Similarly, research on large-scale semantic networks shows that words and concepts in the mental lexicon can be organized as nodes linked through associative relations, enabling efficient retrieval, prediction, and inference (De Deyne et al., 2017). This network representation emphasizes that mental models are not isolated units of knowledge but dynamic, relational structures that support reasoning, learning, and adaptation in complex environments.
Researchers have often used concept mapping as a method to elicit and represent learners’ mental models, particularly in educational contexts. In concept maps, students identify key concepts as nodes and connect them with labeled links that express the relationships, producing a visual network of their understanding. This approach allows teachers and researchers to probe students’ knowledge structures in ways that traditional testing cannot, offering insights into reasoning and misconceptions (Novak & Cañas, 2008). Concept mapping has notable advantages: it makes students’ thinking visible, encourages active learning, and supports metacognitive reflection. However, it also has key limitations. Because students generate their own nodes and links, concept maps are highly variable, which makes grading qualitative and subjective. Additionally, since learners often produce different sets of concepts, it is difficult to make direct comparisons between students’ maps or to benchmark them against expert models (Ruiz-Primo & Shavelson, 1996). As a result, while concept mapping remains a valuable tool for exploring mental models, it poses challenges for standardization and large-scale assessment.
Constrained concept mapping (CCM) is a structured variation of concept mapping in which the instructor provides a predefined set of concepts (nodes) and relationship labels (links), and students are required to represent their understanding by drawing the relationships among those concepts. The advantage of CCM lies in its quantitative grading: similarity scores such as the NetSim (Goldsmith, Johnson, & Acton, 1991) measure calculate the percentage of links shared between a student’s and an expert’s network.
This allows for automated and immediate scoring without the need for instructor grading. Beyond efficiency, CCM enables instructors to aggregate results across students to identify misunderstood concepts, compare individual and group models, and provide targeted remediation. It also offers formative feedback by highlighting missing, extraneous, or mislabeled links. Furthermore, because CCM tasks are generative and allow repeated attempts until mastery, they may engage students more deeply and leverage the testing effect to reinforce learning (Ifenthaler, 2010). As such, constrained mapping combines the cognitive benefits of concept mapping with the practicality of objective grading and adaptive instruction.
Our program, LinkIt, demonstrates the power of constrained concept mapping (CCM) as a tool for eliciting, representing, and assessing both instructor and student mental models. In LinkIt, instructors are first provided with a set of node names and relationship labels to construct an “expert” network that serves as the reference model. Students then use the same nodes and labels to build their own networks and, once satisfied, submit them by clicking a Continue button. The program quantitatively evaluates student performance using the NetSim score, which measures the proportion of links shared between the student and instructor networks (Goldsmith, Johnson, & Acton, 1991). In addition to scoring, LinkIt provides qualitative feedback by overlaying the instructor’s network on top of the student’s. Missing links (errors of omission) appear in blue, and incorrect links (errors of commission) appear in red. Students can click on highlighted links to receive AI-generated explanations of why a link is missing or incorrect, which supports targeted feedback and iterative learning. Using data from anesthesiology concepts, we show how LinkIt can accelerate both learning and assessment—an especially valuable feature for domains like medical training, where rapid acquisition of accurate conceptual knowledge is essential (Ifenthaler, 2010).
References
American Medical Association (2023). Scope of practice: Education matters. Retrieved on 9/16/2025 from https://www.ama-assn.org/practice-management/scope-practice/scope-practice-education-matters.
De Deyne, S., Kenett, Y. N., Anaki, D., Faust, M., & Navarro, D. (2017). Large-scale network representations of semantics in the mental lexicon. The Routledge Handbook of Semantics, 174–202.
Geibel, P. (1998). On the graph-based induction of spatial relations in mental models. Proceedings of the European Conference on Artificial Intelligence.
Goldsmith, T. E., Johnson, P. J., & Acton, W. H. (1991). Assessing similarities of concept maps. Educational Psychology Review, 3(1), 95–111.
Ifenthaler, D. (2010). Relating macro- and micro-level processes of conceptual change by using computer-based learning environments. Computers & Education, 55(2), 429–436.
Jahn, G., Johnson-Laird, P. N., & Knauff, M. (2005). Preferred mental models in spatial reasoning. Cognitive Science, 29(4), 719–765.
Johnson-Laird, P. N. (1983). Mental models: Towards a cognitive science of language, inference, and consciousness. Harvard University Press.
Li, M., & Maani, K. E. (2011). Dynamic decision making: Learning and mental models. Management Learning, 42(2), 195–212.
Novak, J. D., & Cañas, A. J. (2008). The theory underlying concept maps and how to construct and use them. Technical Report IHMC CmapTools 2006-01 Rev 01-2008, Florida Institute for Human and Machine Cognition.
Ruiz-Primo, M. A., & Shavelson, R. J. (1996). Problems and issues in the use of concept maps in science assessment. Journal of Research in Science Teaching, 33(6), 569–600.
Schaffernicht, M., & López‐Astorga, M. (2021). Beyond misperception: Two types of mental model errors in a system dynamics microworld. System Dynamics Review, 37(3), 247–270.
Wykes, C. (1984). Book Review: Mental Models (Johnson-Laird, 1983). Language and Speech, 27(4), 381–385.
Further, once in practice, healthcare professionals continue their education via continuous medical education (CME) credits. These requirements vary by state and certifying body, so details depend on where you practice. For example, Texas MD/DO physicians must do 48 credits every 24 months.
Medical training is also time-consuming and expensive for the instructors who deliver lessons, provide feedback and grade exams. To alleviate some of these inefficiencies, we have been developing a tool based on mental models and concept mapping.
Mental models are internal, simplified representations of a topic that people use to reason, predict, and learn. They enable individuals to simulate outcomes, evaluate alternatives, and guide decision-making. This framework, first formalized by Johnson-Laird, suggests that reasoning does not rely on formal logic but rather on constructing and manipulating these internal models to “read off” possible conclusions (Johnson-Laird, 1983; Wykes, 1984). Mental models are not static; they evolve as individuals incorporate new information, supporting both problem-solving and adaptive learning processes (Li & Maani, 2011). They also enable people to run “mental simulations” of events, which underpin complex tasks such as conditional reasoning, spatial reasoning, and decision-making in uncertain environments (Jahn, Johnson-Laird, & Knauff, 2005). Beyond supporting individual cognition, mental models are dynamic tools for learning—changing existing knowledge structures to accommodate new experiences and helping to avoid errors in judgment that stem from incomplete or flawed internal representations (Schaffernicht et al., 2021).
Mental models can be represented as networks, where nodes correspond to concepts and labeled links capture the relationships among those concepts. This graph-based perspective highlights how knowledge structures are interconnected, allowing individuals to navigate between concepts and reason about their relations. For instance, in spatial reasoning tasks, mental models have been modeled using labeled graphs in which nodes represent objects and edges encode spatial relations between them (Geibel, 1998). Similarly, research on large-scale semantic networks shows that words and concepts in the mental lexicon can be organized as nodes linked through associative relations, enabling efficient retrieval, prediction, and inference (De Deyne et al., 2017). This network representation emphasizes that mental models are not isolated units of knowledge but dynamic, relational structures that support reasoning, learning, and adaptation in complex environments.
Researchers have often used concept mapping as a method to elicit and represent learners’ mental models, particularly in educational contexts. In concept maps, students identify key concepts as nodes and connect them with labeled links that express the relationships, producing a visual network of their understanding. This approach allows teachers and researchers to probe students’ knowledge structures in ways that traditional testing cannot, offering insights into reasoning and misconceptions (Novak & Cañas, 2008). Concept mapping has notable advantages: it makes students’ thinking visible, encourages active learning, and supports metacognitive reflection. However, it also has key limitations. Because students generate their own nodes and links, concept maps are highly variable, which makes grading qualitative and subjective. Additionally, since learners often produce different sets of concepts, it is difficult to make direct comparisons between students’ maps or to benchmark them against expert models (Ruiz-Primo & Shavelson, 1996). As a result, while concept mapping remains a valuable tool for exploring mental models, it poses challenges for standardization and large-scale assessment.
Constrained concept mapping (CCM) is a structured variation of concept mapping in which the instructor provides a predefined set of concepts (nodes) and relationship labels (links), and students are required to represent their understanding by drawing the relationships among those concepts. The advantage of CCM lies in its quantitative grading: similarity scores such as the NetSim (Goldsmith, Johnson, & Acton, 1991) measure calculate the percentage of links shared between a student’s and an expert’s network.
This allows for automated and immediate scoring without the need for instructor grading. Beyond efficiency, CCM enables instructors to aggregate results across students to identify misunderstood concepts, compare individual and group models, and provide targeted remediation. It also offers formative feedback by highlighting missing, extraneous, or mislabeled links. Furthermore, because CCM tasks are generative and allow repeated attempts until mastery, they may engage students more deeply and leverage the testing effect to reinforce learning (Ifenthaler, 2010). As such, constrained mapping combines the cognitive benefits of concept mapping with the practicality of objective grading and adaptive instruction.
Our program, LinkIt, demonstrates the power of constrained concept mapping (CCM) as a tool for eliciting, representing, and assessing both instructor and student mental models. In LinkIt, instructors are first provided with a set of node names and relationship labels to construct an “expert” network that serves as the reference model. Students then use the same nodes and labels to build their own networks and, once satisfied, submit them by clicking a Continue button. The program quantitatively evaluates student performance using the NetSim score, which measures the proportion of links shared between the student and instructor networks (Goldsmith, Johnson, & Acton, 1991). In addition to scoring, LinkIt provides qualitative feedback by overlaying the instructor’s network on top of the student’s. Missing links (errors of omission) appear in blue, and incorrect links (errors of commission) appear in red. Students can click on highlighted links to receive AI-generated explanations of why a link is missing or incorrect, which supports targeted feedback and iterative learning. Using data from anesthesiology concepts, we show how LinkIt can accelerate both learning and assessment—an especially valuable feature for domains like medical training, where rapid acquisition of accurate conceptual knowledge is essential (Ifenthaler, 2010).
References
American Medical Association (2023). Scope of practice: Education matters. Retrieved on 9/16/2025 from https://www.ama-assn.org/practice-management/scope-practice/scope-practice-education-matters.
De Deyne, S., Kenett, Y. N., Anaki, D., Faust, M., & Navarro, D. (2017). Large-scale network representations of semantics in the mental lexicon. The Routledge Handbook of Semantics, 174–202.
Geibel, P. (1998). On the graph-based induction of spatial relations in mental models. Proceedings of the European Conference on Artificial Intelligence.
Goldsmith, T. E., Johnson, P. J., & Acton, W. H. (1991). Assessing similarities of concept maps. Educational Psychology Review, 3(1), 95–111.
Ifenthaler, D. (2010). Relating macro- and micro-level processes of conceptual change by using computer-based learning environments. Computers & Education, 55(2), 429–436.
Jahn, G., Johnson-Laird, P. N., & Knauff, M. (2005). Preferred mental models in spatial reasoning. Cognitive Science, 29(4), 719–765.
Johnson-Laird, P. N. (1983). Mental models: Towards a cognitive science of language, inference, and consciousness. Harvard University Press.
Li, M., & Maani, K. E. (2011). Dynamic decision making: Learning and mental models. Management Learning, 42(2), 195–212.
Novak, J. D., & Cañas, A. J. (2008). The theory underlying concept maps and how to construct and use them. Technical Report IHMC CmapTools 2006-01 Rev 01-2008, Florida Institute for Human and Machine Cognition.
Ruiz-Primo, M. A., & Shavelson, R. J. (1996). Problems and issues in the use of concept maps in science assessment. Journal of Research in Science Teaching, 33(6), 569–600.
Schaffernicht, M., & López‐Astorga, M. (2021). Beyond misperception: Two types of mental model errors in a system dynamics microworld. System Dynamics Review, 37(3), 247–270.
Wykes, C. (1984). Book Review: Mental Models (Johnson-Laird, 1983). Language and Speech, 27(4), 381–385.
Event Type
Poster Presentation
TimeTuesday, March 244:45pm - 6:15pm EDT
LocationRhinelander Gallery
Simulation and Education