Agent-Enriched Data Mining Using an Extendable Framework

An extendable and generic Agent Enriched Data Mining (AEDM) framework, EMADS (the Extendable Multi-Agent Data mining System) is described. The central feature of the framework is that it avoids the use of ontologies or agreed meta-language formats by supporting a system of wrappers. The advantage offered is that the system is easily extendable, further data agents and mining agents can simply be added to the system. A demonstration EMADS framework is currently available. The paper includes details of the EMADS architecture and the wrapper principle incorporated into it. A full description of the framework’s operation is provided by considering two AEDM scenarios; the scenarios are also the focus for an evaluation of the framework. 1. MOTIVATION AND GOALS Agent-Enriched Data Mining (AEDM), also known as multiagent data mining, seeks to harness the general advantageous of MAS in the application domain of Data Mining (DM). MAS technology has much to offer DM, particularly in the context of various forms of distributed and cooperative DM. Distributed (and parallel) DM is directed at reducing the time complexity of computation associated with the increasing sophistication, size and availability of the data sets we wish to mine. Cooperative DM encompasses ensemble mechanisms and techniques such as bagging and boosting. MAS have a clear role in both these areas. MAS technology also offers some further advantageous for AEDM, namely: • Extendibility of DM frameworks, • Resource and experience sharing, • Greater end-user accessibility, • Information hiding, and • The addressing of privacy and security issues. The last of the above advantageous merits some further comment. By its nature DM is often applied to sensitive data. The MAS approach would allow data to be mined remotely. Similarly, with respect to DM algorithms, MAS can make use of algorithms without necessitating their transfer to users, thus contributing to the preservation of intellectual property rights. MAS make it possible for software services to be provided through the cooperative efforts of Cite as: Agent-enriched data mining using an extendable framework, Kamal Ali Albashiri, and Frans Coenen, Proc. of 4th Int. ws on Agents and Data Mining Interaction (ADMI-2009), May, 10–15, 2009, Budapest, Hungary, pp. XXX-XXX. Copyright c © 2008, International Foundation for Autonomous Agents and Multiagent Systems (www.ifaamas.org). All rights reserved. distributed collections of autonomous agents. Communication and cooperation between agents are brokered by one or more facilitators, which are responsible for matching requests, from users and agents, with descriptions of the capabilities of other agents. Thus, it is not generally required that a user or agent know the identities, locations, or number of other agents involved in satisfying a request. The challenge of generic AEDM is the disparate nature and variety of modern DM, and the necessary communication mechanism required to cope with this disparate nature. One approach is to make use of the established Agent Communication Languages (ACLs) and mechanisms; well known examples include the Knowledge Query and Manipulation Language (KQML), the Knowledge Interchange Format (KIF), and the Foundation for Intelligent Physical Agents (FIPA) ACL [14]. All these ACLs have their advantageous and disadvantageous and tend to address particular forms of intra-agent communication; for example FIPA ACL is directed at agent negotiation. Each can be employed in the context of AEDM communication but on its own will not facilitate the shared agent understanding required to achieve generic AEDM. This would require recourse to the use of ontologies and/or some agreed metalanguage. It is suggested in this work that a method of addressing the communication requirements of AEDM is to use a system of mediators and wrappers coupled with an ACL such as FIPA ACL, and that this can more readily address the issues concerned with the variety and range of contexts to which AEDM can be applicable. To investigate and evaluate the expected advantageous of wrappers and mediators, in the context of generic AEDM, the authors have developed and implemented (in JADE) a multi-agent platform, EMADS (the Extendable Multi-Agent Data mining System). Extendibility is seen as an essential feature of the framework primarily because it allows its functionality to grow in an incremental manner. The vision is of an “anarchic” collection of agents, contributed to by a community of EMADS users, that exist across an “internet space”; that can negotiate with each other to attempt to perform a variety of DM tasks (or not if no suitable collection of agents come together) as proposed by other (or the same) EMADS users. An EMADS demonstrator is currently in operation. The primary goal of the EMADS framework is to provide a means for integrating new DM algorithms and data sources in a distributed infrastructure and collaborative environment. However, EMADS also seeks to address some of the issues of DM that would benefit from the rich and complex interactions of communicating agents. The broad advantages offered by the framework are: • Flexibility in assembling communities of autonomous service providers, including the incorporation of existing applications. • Minimization of the effort required to create new agents, and to wrap existing applications. • Support for end users to express DM requests without having detailed knowledge of the individual agents. The rest of this paper is organised as follows. A brief review of some related work on Agent-enriched Data Mining (AEDM) is presented in Section 2. The conceptual framework together with an overview of the wrapper principle is presented in Section 3. The framework operation is illustrated in Section 4 using two DM scenarios: Meta Association Rule Mining (MARM) and single label classification. Finally some conclusions are presented in Section 5.