How to share experience and resources among learners is becoming one of the hottest topics in the field of E-Learning collaborative techniques. An intuitive way to achieve this objective is to group learners which can...How to share experience and resources among learners is becoming one of the hottest topics in the field of E-Learning collaborative techniques. An intuitive way to achieve this objective is to group learners which can help each other into the same community and help them learn collaboratively. In this paper, we proposed a novel community self-organization model based on multi-agent mechanism, which can automatically group learners with similar preferences and capabilities. In particular, we proposed award and exchange schemas with evaluation and preference track records to raise the performance of this algorithm. The description of learner capability, the matchmaking process, the definition of evaluation and preference track records, the rules of award and exchange schemas and the self-organization algorithm are all discussed in this paper. Meanwhile, a prototype has been built to verify the validity and efficiency of the algorithm. Experiments based on real learner data showed that this mechanism can organize learner communities properly and efficiently; and that it has sustainable improved efficiency and scalability.展开更多
This paper presents a new look on emergence from the aspect of locality andglobality of evaluation functions for solving traditional computer problems. We first translate theConstraint Satisfaction Problem (CSP) into ...This paper presents a new look on emergence from the aspect of locality andglobality of evaluation functions for solving traditional computer problems. We first translate theConstraint Satisfaction Problem (CSP) into the multi-agent system, and then show how a globalsolution emerges from the system in which every agent uses a local evaluation function to decide itsaction, while comparing to other traditional algorithms, such as Local search and SimulatedAnnealing which use global evaluation functions. We also give some computer experimental results onlarge-scale N-queen problems and κ-Coloring problems, and show that emergence only depends onproblem instance, not details of agent settings, i.e. in some CSPs, the system can self-organize toa global solution, but can not in some other CSPs no matter what settings of agents have.展开更多
文摘How to share experience and resources among learners is becoming one of the hottest topics in the field of E-Learning collaborative techniques. An intuitive way to achieve this objective is to group learners which can help each other into the same community and help them learn collaboratively. In this paper, we proposed a novel community self-organization model based on multi-agent mechanism, which can automatically group learners with similar preferences and capabilities. In particular, we proposed award and exchange schemas with evaluation and preference track records to raise the performance of this algorithm. The description of learner capability, the matchmaking process, the definition of evaluation and preference track records, the rules of award and exchange schemas and the self-organization algorithm are all discussed in this paper. Meanwhile, a prototype has been built to verify the validity and efficiency of the algorithm. Experiments based on real learner data showed that this mechanism can organize learner communities properly and efficiently; and that it has sustainable improved efficiency and scalability.
基金This paper is supported by the International Program of Santa Fe Institute and the grant of China National Science Foundation(No.70171052).
文摘This paper presents a new look on emergence from the aspect of locality andglobality of evaluation functions for solving traditional computer problems. We first translate theConstraint Satisfaction Problem (CSP) into the multi-agent system, and then show how a globalsolution emerges from the system in which every agent uses a local evaluation function to decide itsaction, while comparing to other traditional algorithms, such as Local search and SimulatedAnnealing which use global evaluation functions. We also give some computer experimental results onlarge-scale N-queen problems and κ-Coloring problems, and show that emergence only depends onproblem instance, not details of agent settings, i.e. in some CSPs, the system can self-organize toa global solution, but can not in some other CSPs no matter what settings of agents have.
