Multi-agent Instructional Communities: A Computational and Experimental Approach

Web为成千上万的社区提供了成千上万的资源。然而，这些资源难以定位且无法协调，从而导致用户的非生产性体验。pi提出了一个基于代理的框架来推理和检索适当的材料，提供有意义的用户指导。这些代理将在教学技术领域，并将采用用户建模和规划来提供健壮的性能，尽管资源的可用性不断变化，网络配置的修改和用户需求的变化。智能体将相互沟通并选择专门的学习资源来解决更大的学习问题的各个方面。他们将协商在可用资源中分配学习任务，确保以协调的方式解决总问题的所有方面。一个模拟的大型在线学习社区将被构建来演示预期的行为。模拟将简化学习和辅导，但将包括解决冲突，并提出避免失败的方法。它将说明综合社区的定性、定量和视觉上的好处，探索在线社区的特性，预测和避免大规模在线学习社区发展中出现的问题。pi将展示哪些技术是有效的，哪些是不切实际的，哪些在大规模网络中是可扩展的。

The Web provides thousands of resources for thousands of communities. However, these resources are difficult to locate and do not coordinate, resulting in unproductive experiences for users. The PIs propose an agent-based framework to reason about and retrieve appropriate materials, providing meaningful user guidance. These agents will be in the area of instructional technology and will employ user modeling and planning to provide robust performance despite changing availability of resources, modifications in the network configuration and variable user needs. Agents will communicate and select specialized learning resources to solve aspects of larger learning problems. They will negotiate distribution of learning tasks among available resources ensuring that all aspects of the total problem are addressed in a coordinated manner. A simulated large-scale on-line learning community will be constsructed to demonstrate the intended behavior. The simulation will simplify learning and tutoring yet will include conflict resolution and propose methods to avoid failure. It will illustrate qualitative, quantitative and visual benefits of the integrated community and explore the properties of on-line communities, anticipating and avoiding problems that arise as massive on-line learning communities develop. The PIs will demonstrate which techniques are productive, which impractical and which are scaleable in a massive network.