Study And Implementation Of Agent-based Parallel Group Data Mining Model

With the universal applications of information technology and popularity of Internet, distributed applications and research of information system have gradually become hot topics since late 20th century. Distributed data mining also attracts more and more attention of researchers. Meanwhile, researches on distributed data mining model, distributed data mining algorithm and distributed data mining tool have also been done and obtained some achievements. However, there also exist many problems to be solved. Multi-Agent technology is considered as an effective method to solve distributed data mining. In distributed data mining model, data mining researchers put forward several classic distributed data mining models based on Agent, such as, JAM, PADMA, as well as data mining system-BODHI based on heterogeneous sites, which improves on the basis of CDM model. These models play a positive role in promoting researches on distributed data mining.Along with the development of the Internet covering the whole world, people have realized long-distance network communication. New features appear in distributed applications: huge number of data stations, expanding new data sites at any time, high frequency of data update, spanning large distance. How to carry out data mining effectively about such a distributed system becomes an urgent task. It is incompetent for classical model in dealing with these new patterns of distributed system.Based on Multi-Agent technology, this thesis analyzes advantages and disadvantages of classic distributed data mining based on Agent. Through improving the classical models, we provide a new parallel groups data mining model PADMAN, which based on Agent and network, and it being use of different network technologies as a communications media. This model aims to meet the demand of the current appearance of huge number of data stations, flexible appearance of new data stations, the high frequency of data update, distributed system data mining which spans large distance. Then, we propose a merger strategy of data mining results being suitable for PADMAN model. Through merged two data mining results, the disadvantages of the home site's big load pressure, large amount of network communication and the higher difficulty of control appearing former models are solved.Based on the researches on PADMAN model framework and the merger strategy, we use Eclipse and the JADE platform to realize the PADMAN prototype system for validating the model of collaborative relationships within the group.This thesis focus on PADMAN model architecture design and merger strategy of data mining results for this model. However, there are some shortcomings in model realizations remains further study. Some new ideas proposed in the paper has certain reference value in coping with new features of distributed data mining.