Inference Engines Performance in Reasoning Tasks for Intelligent Tutoring Systems

Abstract

The use of formal knowledge representation models in intelligent tutoring systems often requires logical reasoning on these models by predefined rules. This process can be time and memory consuming, so finding effective software reasoners for different applications is an important research field. This problem is relevant for cognitive and constraint-based intelligent tutoring systems. We performed a comparative study of various software reasoners (Pellet, Apache Jena inference subsystem, Apache Jena SPARQL query processor, SWI-Prolog with semweb package, Closed World Machine, and Answer Set Programming solvers Clingo and DLV) for solving tasks specific to intelligent tutoring systems using three formal models with different properties and corresponding rule sets created for intelligent tutoring systems in introductory programming education domain. We compared features of rule-definition formalisms for different approaches and measured run and wall time, average CPU load, and peak RAM usage based on the count of inferred RDF triples. The experiments show that Apache Jena infers the solution quicker than other reasoners on the majority of tasks but consumes a significant amount of memory, while Clingo performs significantly better for combinatorial problems.