Intelligent Agent for Delivering Learning Materials Department of Computer Science and Engineering University of Nebraska Co-Sponsored by Great Plains Software Technology Initiative National Center for Information Technology in Education
INTRODUCTION VisionBuild an intelligent agent with machine learning capabilities to deliver better learning materials to students –Employs sound artificial intelligence (AI) techniques such as case-based reasoning (CBR), reinforcement learning, dynamic profiling, semantic search, rule- based reasoning, simulated annealing, and so on –Incorporates instructional technology techniques such as adaptive quiz, learning objects, learner modeling, and so on –Investigates how agents can learn to deliver better learning materials to students
Initial Goals and Objectives To build an agent capable of adapting to students’ real-time behavior and historical profile for its delivery of examples and exercise problems, of learning useful delivery strategies, and of self-monitoring and evaluation To develop courseware in Discrete Structures and Software Engineering, two low-level undergraduate classes where the class size is usually large and diversity is high To establish a flexible, easy-to-use database of courseware and agent execution, from operational items such as student profiles, agent success rates, etc., to educational items such as learner model, domain expertise, and course content To evaluate the usefulness of agent as (a) an on-line Teaching Assistant for classes in Discrete Structures and Software Engineering, (b) a distance course agent, and (c) a testbed for analyzing instructional designs and theories
PERSONNEL Leen-Kiat Soh Suzette Person LD Miller Todd Blank Ashok Kumar Thirunavukkarasu Brandi Hobbs (Spring 2003)
OVERVIEW ILMDA Reasoning student Computer & GUI database lectures Historical profile, Real-time behavior Parametric profile of student and environment Retrieval instructions Profile updates Statistics updates Timely delivery of examples & exercise problems Examples Exercise problems Statistics ILMDA Agent
PHASE I: Getting Started Started in February 2003 Ended with a Report on June 16 th and a Demo to Rich Sincovec –GUI to interface with students –CBR modules basically completed –Web-based content entry started –Learner modeling completed –Basic agent completed –Databases (mySQL) organized –Ideas and concepts fleshed out
PHASE I: Getting Started Databases to keep track of students, learning materials, and run-time info
PHASE II: Getting Smart! Started in June 2003 Ended with a Report on August 28 th and a Demo today on September 10 th –A knowledge engineering GUI tool for adaptation heuristics and graph semantics –A graph parser and a heuristics parser –The syntactic definition of adaptation heuristics –The syntactic definition of graph semantics –A graph traversal module for adaptation in case-based reasoning –A simulated annealing module for adaptation in case-based reasoning (conceptual fleshout) –A usage history profiler of cases –A content entry GUI for adding tutorial materials to the database
PHASE II: Getting Smart! Knowledge Engineering GUI Content Entry GUI Tutorial Database Adaptation Heuristics Heuristics Parser Graph Semantics Graph Parser CBR: Adaptation Graph Traversal ILMDA Agent Simulated Annealing Relaxed Retrieval Casebase Usage History Profiler
PHASE II: Graph Maker
PHASE II: Adaptation Heuristics
Line 2, DIFFICULTYLEVEL: GPA – If a student has a higher GPA, then the agent recommends a more difficult problem or example. 8 Successes – If a student has had more successes, then the agent recommends a much more difficult problem or example. 2 numSessions – If a student has taken more sessions, then the agent recommends a more difficult problem or example. 1 AveGrade – If a student has a higher grade in problem solving, then the agent recommends a much more difficult problem or example. … -1 aveExmpToTtrl/aveProbToExmp/aveProbToTtrl – If a student moves back and forth often between learning items (examples, problems, tutorial), then the agent recommends an easier problem or example.
PHASE II: Interest Graph
PHASE II: Simulated Annealing
PHASE III: Becoming Useful Sep 10Prototype Demo II to Rich Sincovec, Roger Bruning, Art Zygielbaum Sep 20Explore partnership with Gallop Research Center (Contacts made already) Sep 30Simulated Annealing and Usage History Profile completed Dec 1Revised rules/heuristics from the educational researchers obtained and implemented in the agent Dec 1Partnership with Gwen Nugent to export our agent technology to BRIDGES Dec 1Demonstrate a prototype using relaxed retrieval scheme for our database Master’s Project Defense for Ashok Kumar Thirunavukkaras Dec 1Demonstrate a prototype using good, solid set of courseware materials Report on Phase III
DEMO & INFO