SOFTWARE AGENTS IN E-LEARNING SYSTEMS

Abstract

The usage of the multi-agent approach for solving resource management problems in e-learning systems is discussed in this paper. The advantages of multi-agent system comparing with traditional systems are emphasized. Such systems provide simplicity of implementation, transferability, and scalability, allow parallel computing, are guided by systems based on knowledge, have the ability to self-organize and evolve. The formalization of dynamic distribution of resources based on the network of the needs and opportunities is proposed. Two types of autonomous interacting agents are identified. They need agents and opportunity agents. All agents act in accordance with their own goals and according to certain rules, allowing them to act independently and interact with each other. As a relation between two types of agents was used conformity. The basic principles of network construction are defined. The formal model of interaction between agents is described. It was defined that one of the most often problems that can arise in the e-learning system is the problem of coordination. Partly this problem can be solved by the detailed elaboration of a set of decision-making rules. The article proposes an algorithm that allows flexible and prompt solving of information resources distribution tasks in accordance with the needs of rapidly changing environments. The specialized components for the agents in the system are defined: the agent communication component, the decision making component, and the component of the flow calculations. The main obstacles for multi-agent approach implementation are defined. They are: difficulties to evaluate the optimality of the decision; solutions are sensitive to the history of events; small changes in the system input can lead to significant changes in the output; there are some difficulties in adjusting the solution in the "manual" mode; misunderstandings in explaining the results as a consequence of complex causal relationships are possible; delays of solutions are liable to occur because of long chains of changes; it is possible to obtain non-identical solutions under the same conditions of input when re-launching the model.