关联规则挖掘是数据挖掘中成熟的技术之一，在商业管理、政府办公、科学研究和工程开发中得到了广泛应用。但其处理的目标常常是大规模的数据集，处理的是异构类型数据，另外考虑到数据挖掘的安全性、容错性等问题，为此迫切需要一种手段能够智能地、有效地、安全地挖掘出事务数据间有趣的关联规则。而分布式人工智能的前沿技术Multi-Agent具有高度智能化、易于构造分布式系统和软件复用性强等优点，这为关联规则挖掘提供了强有力支持。本文介绍了关联规则挖掘技术和Multi-Agent技术，并把它们应用到一个行为推荐原型系统当中，设计了一个基于Multi-Agent和关联规则挖掘的ARSM系统(The Action Recommendation Prototype System for User Based on Multi-Agent and Association Rule Mining)，并将其应用到Web访问上。本文的主要工作如下： 利用已存在的BFP-Miner算法中改进的FP-Tree构造方法和基于位对象的频繁k-项集挖掘方法挖掘包含频繁k-项集的频繁(k+1)-项集，再根据最小置信度产生强关联规则。在ARSM系统中，将此产生强关联规则的方法应用于对用户行为日志进行挖掘，以产生行为推荐。 通过对系统推荐任务的规划和分解，设计了ARSM系统。ARSM系统由UserAgent、DProAgent、ARAgent和Action Log Data Base四部分组成，其中ARAgent负责管理DatabaseProcessAgent、BFPTreeMinerAgent和ARMAgent三个Agent。为实现每个Agent所具备的功能，设计了Agent的模型和结构、说明了Agent的工作流程及控制算法、定义了每个Agent具备的技能。通过Multi-Agent环境下Agent之间交互细节的分析，给出了ARSM系统中Multi-Agent的管理结构图。 把ARSM系统应用到Web访问上，为Web访问者进行行为推荐。实验中采用了来自于微软网站的匿名网络数据集，对它进行了预处理，然后在这个数据集上实现对Web访问者进行行为推荐的系统目标。文章说明了对Web访问者进行行为推荐的系统任务的实现过程，并对推荐过程中Agent主要技能给出实现，最后对系统的实验效果做了说明和分析。

Association rules mining is one of the most mature technology in data mining. It has been applied widely in business management, government office, scientific research and engineering development. Because the processing datasets are usually large- scale and heterogeneous ones, and considering the security and fault tolerance etc, a method which can mine interest association rules from the database intellectually, effectively, safely is needed urgently. Because of the high intelligence, effectiveness of constructing distributing system, and powerful reusable features of the Multi-Agent technology which is the top technology of distributed artificial intelligent, it provides strong support for association rule mining etc. The technology of association rule mining and the Multi-Agent are described in this paper. A system of ARSM based on association rule mining and Multi-Agent is designed, and it is applied to the action recommendation system of Web visiting. The main work is stated as follows: Firstly, using the improved constructing method of FP-Tree (Frequent Pattern tree) and the mining method of the frequent k-item set based on bit objects in BFP-Miner(Bit Frequent Pattern Miner) algorithm that has been proposed, the frequent k+1-item set that includes the frequent k-item set is generated, then the strong association rules is obtained according to the min confidence. In the system of ARSM, the method that generates the strong association rules is applied to mine the Web log to recommend action for Web visitor. Secondly, the system ARSM has been designed through analyzing the task of it. The architecture of ARSM is composed of four parts which are UserAgent、 DProAgent、ARAgent and Action Log Data Base. ARAgent manages the three Agents which are DatabaseProcessAgent、BFPTreeMinerAgent and ARMAgent. In order to realize the function of each Agent, the model and the architecture of the Agent have been designed, the working process and control algorithm is also introduced. The skills of every Agent are defined. In the end, a manage framework of ARSM based on Multi-Agent is provided by analyzing the communication among Agents. Finally, by pre-proceeding the anonymous Web data that comes from the Web of Microsoft, ARSM is applied to the Web visiting to recommend action for the user. The recommend process is implemented and the skills of Agent is realized. In the end, the experimental result of the system is displayed to demonstrate its effectiveness.