Context. Our schools should practice individualized project based education, in which students learn by solving real-world problems. However, widespread individualized project based education is currently infeasible because of the demands it places on teachers. Comprehensive infrastructure to support project based education would include a "network of ideas" with multimedia presentations to convey and stimulate ideas for projects, and an intelligent expert system for substantive support of project definition: the selection and planning of projects.
Project goals. 1) Design and implement knowledge representation for project definition, including the ability to identify projects which are appropriate given students' previous project experience, and to inform decisions for choosing between options for implementing a project. 2) Evaluate artificial intelligence classification taxonomies as the technological basis for an expert system to support project definition.
Design. Projects in a project web are composed of logical groupings of subprojects. Subject matter content is represented with a subsumption lattice of domain concepts. Projects are related to each other via composition and through their links to the domain lattice. Domain overlap between sets of projects is quantified as a function of the number of shared domain concepts, their density, and the directness of their connection to the projects.
Results. CLASSIC, a description logic system in the KL-ONE family, was used as the basis for an initial implementation. A test knowledge base with 37 projects and 98 domain concepts was developed. A general function to evaluate the proximity of sets of projects is used in a variety of ways to illustrate its usefulness and limitations for planning.
Duration layer: WEEKS-LONG Student experience: (EXPERIMENT-DOG-BODY-LANGUAGE EXPERIMENT-HUMAN-NON-VERBAL-COMMUNICATION) Shared Overlap Project Concepts Density Strength Summary ----------------------------------------- -------- -------- -------- -------- EXPERIMENT-FEEDING-SQUIRRELS 17 0.7391 0.2666 3.3498 EXPERIMENT-BIRD-INTERACTION-WHILE-FEEDING 16 0.7273 0.2781 3.2362 EXPERIMENT-FIGHTING-FISH-DISPLAYS 16 0.6957 0.2685 2.9888 EXPERIMENT-MARE-AND-FOAL-INTERACTION 7 0.2333 0.1614 0.2636 EXPERIMENT-MOMMY-DUCK-AND-DUCKLINGS 4 0.1538 0.1340 0.0825 ESTIMATE-STAR-DISTANCE 1 0.0323 0.0824 0.0027 SURVEY-FARM 0 0.0000 0.0000 0.0000
Conclusions. Subsumption inheritance is appropriate for structuring the domain lattice. But, CLASSIC's inability to model dynamic or disjunctive data in an integrated manner makes it inadequate for further development of the system, which will extend behavior requiring represention of students and planning activity. The current exercise also illuminates issues related to the separation of concepts and individuals in classification taxonomies; the use of proximity in a semantic network as a means for judging similarity; and the challenge of accurately expressing the meaning of composition links between projects and their subprojects.