Ontology & Knowledge Base

Understanding adaptive learning requires familiarity with its specialized terminology and conceptual structures. This ontology provides precise definitions of key terms, taxonomies of system types, and a framework for understanding how the pieces fit together.

Core Concepts

Knowledge Component (KC)

A Knowledge Component is an atomic piece of knowledge or skill that can be assessed and learned. KCs form the basic units that adaptive systems track. Examples include "solving linear equations," "identifying verb tenses," or "applying Ohm's law."

Knowledge Components are also called:

Skills
Competencies
Learning objectives
Concepts

Knowledge State

The Knowledge State represents what a learner currently knows and doesn't know, typically expressed as a vector of probabilities (one per KC) or as parameters in a more complex model. The adaptive system's goal is to maintain an accurate estimate of the learner's knowledge state.

Learning Path

A Learning Path is a sequence of learning activities (content items, exercises, videos) designed to help a learner achieve learning objectives. Adaptive systems dynamically construct personalized learning paths based on learner state and goals.

System Taxonomy

By Adaptation Target

Type	What Adapts	Example
Content-level	Which content is shown	Different next problem based on performance
Presentation-level	How content is presented	Video vs text vs interactive simulation
Sequence-level	Order of content	Prerequisites enforced, optimal path calculated
Scaffolding-level	Support provided	Hints, worked examples fade as skill develops

By Technical Approach

Rule-based: Uses expert-defined if-then rules for adaptation
Model-based: Uses mathematical models of learning (e.g., BKT)
Data-driven: Uses machine learning on historical data
Hybrid: Combines multiple approaches

By Domain

Well-defined domains: Mathematics, physics, programming with clear right/wrong answers
Ill-defined domains: Writing, design, critical thinking with subjective evaluation
Procedural domains: Skills requiring step-by-step execution
Declarative domains: Facts, concepts, and principles

Glossary of Terms

A

Adaptive Assessment: Testing that adjusts difficulty based on student responses to efficiently estimate ability level.

Affective Computing: Systems that recognize and respond to learner emotions and engagement states.

Auto-Encoder: Neural network architecture used in some knowledge tracing approaches for dimensionality reduction.

B

Bayesian Knowledge Tracing (BKT): A probabilistic model using Hidden Markov Models to track mastery of individual knowledge components.

Bloom's 2 Sigma Problem: The finding that one-on-one tutoring produces learning outcomes two standard deviations better than conventional instruction.

C

Cognitive Load: The mental effort required to process information, which adaptive systems aim to optimize.

Cognitive Tutor: Intelligent tutoring systems based on cognitive architecture models like ACT-R.

Cold Start Problem: The challenge of making accurate predictions for new users or items without sufficient historical data.

Computerized Adaptive Testing (CAT): Assessment systems that select questions based on previous responses to efficiently measure ability.

Constraint-Based Modeling (CBM): An approach that represents domain knowledge as constraints rather than procedures.

D

Deep Knowledge Tracing (DKT): Knowledge tracing using recurrent neural networks, particularly LSTMs.

Desirable Difficulty: The principle that some learning challenges improve long-term retention.

Difficulty Parameter: In IRT models, the ability level at which a learner has 50% probability of answering correctly.

E

Educational Data Mining (EDM): The application of data mining techniques to educational data.

Engagement: The degree of attention, curiosity, interest, optimism, and passion that learners show.

Expert System: AI systems that emulate the decision-making ability of a human expert.

F

Fading: Gradually removing scaffolding or hints as learner competence increases.

Feature Engineering: The process of selecting and transforming variables for use in machine learning models.

G

Gamification: Application of game-design elements in non-game contexts to increase engagement.

Granularity: The level of detail at which knowledge components are defined.

H

Hidden Markov Model (HMM): Statistical model used in BKT where the learner's true knowledge state is hidden but inferred from observations.

I

Intelligent Tutoring System (ITS): Computer systems that provide immediate and customized instruction or feedback to learners.

Item Response Theory (IRT): A statistical framework relating latent traits to observable test responses.

K

Knowledge Graph: A network representation of domain knowledge with concepts as nodes and relationships as edges.

Knowledge Space: A mathematical structure representing the possible states of knowledge in a domain and feasible learning paths.

Knowledge Tracing: The problem of modeling a student's knowledge over time based on their interaction history.

L

Learning Analytics: The measurement, collection, analysis and reporting of data about learners and their contexts.

Learning Management System (LMS): Software application for the administration, documentation, tracking, and delivery of educational courses.

Learning Object: A digital resource that can be reused to support learning.

LSTM (Long Short-Term Memory): A type of recurrent neural network used in DKT and other sequence models.

M

Mastery Learning: An instructional strategy where students must achieve a defined level of mastery before progressing.

Metacognition: Awareness and understanding of one's own thought processes.

MOOC: Massive Open Online Course, often generating large datasets for adaptive learning research.

O

Open Learner Model: Visualization of the system's model of the learner's knowledge, shown to the learner.

P

Personalized Learning: Instruction paced to learning needs and tailored to preferences and interests.

Prerequisite: A knowledge component that must be mastered before another can be learned effectively.

Q

Q-Matrix: A binary matrix mapping assessment items to the knowledge components they test.

R

Reinforcement Learning (RL): Machine learning where an agent learns by interacting with an environment and receiving rewards.

Retrieval Practice: Learning strategy involving recalling information from memory, often used in adaptive systems.

S

Scaffolding: Temporary support structures that help learners accomplish tasks they cannot yet do independently.

Spaced Repetition: Learning technique with review intervals increasing over time to optimize retention.

Student Model: The component of an ITS that represents what the system believes about the learner's state.

T

Transfer Learning: Applying knowledge learned in one context to new, similar contexts.

U

User Model: See "Student Model" or "Learner Model."

Z

Zone of Proximal Development (ZPD): The difference between what a learner can do independently and what they can do with guidance.

Model Relationships

Adaptive learning systems typically involve several interconnected models:

Domain Model: What is to be learned (KCs, prerequisites, knowledge graph)
Learner Model: What the student currently knows (knowledge state)
Tutoring Model: How to teach (content selection, feedback strategies)
Interface Model: How to present content and interact with learners

These models work together: the Domain Model defines the space of possible knowledge, the Learner Model tracks position in that space, the Tutoring Model decides how to navigate, and the Interface Model executes the interaction.