Electronic Business: Concepts, Methodologies, Tools, and Applications (4-Volumes) P86 pot

784 Applying Information Gathering Techniques in Business-to-Consumer and Web Scenarios • Autonomy: agents work without the direct intervention of humans or others, and have some kind of control over their actions and internal state. • Social ability: agents interact or commu- nicate with other agents. • Reactivity: agents perceive their environ- ment (which may be the physical world, a user via a graphical user interface, a collection of other agents, the Internet, or perhaps all of these combined), and responds in a timely fashion to changes that occur in it. • Pro-activity: agents do not simply act in response to their environment, they are able to exhibit goal-directed behaviour by taking the initiative. An agent is capable of handling complex, high-level tasks. The decision as to how such tasks are best split up into smaller sub-tasks, and in which order and way the sub-tasks are best performed, should be made by the agent itself. • Temporal continuity: agents are continu- ously running processes. • Mobility: an agent has the ability to transport itself from one machine to another, retaining its current state. An MAS is a system composed of a popula- tion of autonomous agents, which cooperate with each other to reach common objectives, while simultaneously pursuing individual objectives (Wooldridge, 2002). In order to solve common problems coherently, the agents must commu- nicate amongst themselves and coordinate their activities. Coordination and communication are FHQWUDOWR0$6 IRUZLWKRXWLWDQ\EHQH¿WVRI interaction vanish and a group of agents quickly degenerates into a collection of individuals with chaotic behaviour. In the most general case, agents will be acting on behalf of users with different goals and motivations. The characteristics of any MAS can be summarized as follows (Sycara, 1998): • every agent has incomplete information or capabilities for solving the problem and, thus, has a limited viewpoint; • there is no global control; • data are decentralized; • computation is asynchronous. Web Services Technologies Web services have emerged as the next generation of Web-based technology for exchanging informa- tion. Web services are modular, self-describing, self-contained applications that are accessible over the Internet. Based on open standards, Web services allow the implementation of Web-based applications using any platform, object model, or programming language. Web services are services offered via the Web. In a typical Web-services scenario, a busi- ness application sends a request to a service at a given URL using the SOAP (Simple Object Access Protocol) protocol over HTTP, and uses XML (eXtensible Markup Language) as the base language. The service receives the request, pro- cesses it and returns a response. A well-known example of a Web service is that of a stock quote service (i.e., a book store), in which the request DVNVIRUWKHFXUUHQWSULFHRIDVSHFL¿HGERRNLQ stock, and the response gives the price. This is one of the simplest forms of a Web service in WKDWW KHUHTXHVWLV¿OOHGDOPRVWLP PHGLDWHO\Z LWK the request and response being parts of the same method call. Where the current Web enables users to con- nect to applications, the Web-services architecture enables applications to connect to other applica- tions. Web services is therefore a key technology in enabling business models to move from B2C (Business to Consumer) to B2B (Business to Business). An enterprise can be the provider of Web services and also the consumer of other Web Services. Also, Web Services are based on a set RIVWDQGDUGL]HGUXOHVDQGVSHFL¿FDWLRQVPDNLQJ it more portable. The generic Web-Services ar- 785 Applying Information Gathering Techniques in Business-to-Consumer and Web Scenarios chitecture is shown in Figure 1, this architecture shows the infrastructure required to support Web services in terms of three roles: service provider, service requestor and service registry, and the necessary transactions and interactions EHWZHHQWKHPSXEOLVK¿QGDQGELQG$service provider publishes a service description to a service registry, a service requesterWKHQ¿QGV the service description via the service registry. Finally, the services description contains suf- ¿FLHQWLQIRUPDWLRQIRUWKHVHUYLFHUHTXHVWRUWR bind to the service provider to use the service (Berners-Lee, 2003; Gardner, 2001). Binding is the process that allows an application to connect to a Web service at a particular Web location and start interacting with it. Using Web services there are three important technologies that are necessary to be familiar with (Gardner, 2001): • Web Services Description Language (WSDL): WSDL is the metadata language of :HE V H U Y L F H V  ,W D F W V D V D ³ X V H U ¶VP D Q X D O ´ IR U  :HEVHUYLFHVGH¿QLQJKRZVHUYLFHSURYLG- ers and requesters communicate with each other about Web services. Similar to XML, WSDL is extensible to allow the description of endpoints and their messages, regardless of what message formats or network proto- cols are used for communicating. Typically, if somebody wants to create an application that communicates with a particular Web service, it is only necessary to describe that VHUYLFH¶V:6'/LQD¿OH:6'/ • Simple Object Access Protocol (SOAP): SOAP is an XML-based protocol for ex- changing information in a decentralized, GLVWULEXWHGHQYLURQPHQW,WGH¿QHVDPHFKD- nism to pass commands and parameters between clients and servers. Like Web services as a whole, SOAP is independent Figure 1. Web Services architecture 786 Applying Information Gathering Techniques in Business-to-Consumer and Web Scenarios of the platform, object model, and program- ming language being used (SOAP, 2003). • Universal Description, Discovery, and Integration (UDDI): UDDI is the meet- ing place for Web services. An information database of Web services, a UDDI registry stores descriptions about companies and the services they offer in a common XML format. Web-based applications interact with a UDDI registry using SOAP mes- sages. Conceptually, the data in a UDDI registry can be divided into three different types of directories: a white-pages section that provides business contact information, a yellow-pages section that categorizes businesses and services, and a green-pages section that provides technical information about the services that a business offers (OASIS-UDDI, 2005). RELATED WORK This section provides a brief description of several Web Information Gathering Systems that cur- rently have been designed and deployed to work with the information stored in the Web. These systems use different CI techniques and Web technologies to integrate the retrieved informa- tion. The following is a brief description of some of those systems: • WebPlan (Hüllen et al., 1999): is a Web DVVLVWDQWIRUGRPDLQVSHFL¿FVHDUFKRQWKH Internet that is based on dynamic planning and plan execution techniques. The existing planning system CAPlan has been extended in different ways in order to deal with in- complete information, information seeking operators, user interaction, and interleaving planning and execution. WebPlan special- L]HVLQ¿QGLQJVSHFL¿F3&VRIWZDUHRQWKH Internet. Planning is used in this system to select the most appropriate sources to look for information. • SIMS/Ariadne (Knoblock et al., 2000). This system includes a set of tools to construct wrappers that make Web sources look like relational databases. Planning and media- tion techniques are used, both to access the distributed information, and to integrate the information found. • Heracles (Ambite, Barish, Knoblock, Mus- lea, Oh, & Minton, 2002). This framework is used to develop different information assistant systems that employ a set of infor- mation agents (Ariadne, Theseus, Electric Elves). A dynamic hierarchical constraint propagation network (CPN) is used to inte- grate the different information sources. Two assistant systems have been implemented: The Travel Planning Assistant (specialized in assisting tourists to plan their trips) and 7KH:RUOG,QIR$VVLVWDQWIRUDXVHUVSHFL¿HG location, the system integrates information from different information sources such as weather, news, holidays, maps, airports, etc.). In this framework the integration of the retrieved information is made by a CPN. • Argos (Dynamic Composition of Web Ser- vices for Goods Movement Analysis and Planning). The Argos project is developing a new approach to automatic generation of VFLHQWL¿FZRUNÀRZVEDVHGRQ:HEVHUYLFHV There are three main objectives: • To advance computer science research by developing an expressive Web services description language and techniques for dynamically composing Web services. • To develop and conduct test applica - tions of an intra-metropolitan goods PRYHPHQWÀRZPRGHOXVLQJ:HEVHU- vices in cooperation with government partners. 787 Applying Information Gathering Techniques in Business-to-Consumer and Web Scenarios • To use the model to conduct social science research on intra-metropolitan economic linkages and spatial struc- ture. Although the focus is on the specif- ic topic of urban goods movement, the approach to Web service composition is general and can be applied to other VFLHQWL¿FGDWDJDWKHULQJDQGDQDO\VLV tasks. (Ambite et al., 2002; Ambite et al., 2004) • SAMAP (Multiagent Context-sensitive Adaptive Planning System). The main objec- tive of SAMAP is the analysis, design and implementation of a multi-agent system with the ability to perform hierarchical, temporal, and resource planning and scheduling in the area of ubiquitous computing. The system will also be dynamic in that it will be able to learn from past problem solving experi- ences, as well as automatically acquiring a user model. • Intelligent Travel Recommendation (ITR) (Ricci et al., 2002). ITR is a Web-based recommender system that enables the user to select travel locations, activities and attractions, and supports the building of a personalized travel plan. In this system the user is asked explicitly about his or her needs and constraints. The system, combin- LQJ FRQWHQWEDVHG ¿OWHULQJ WHFKQRORJLHV interactive query management, and varia- WLRQVRIWKHFROODERUDWLYH¿OWHULQJDSSURDFK or case-based reasoning, ranks suggestions extracted from structured catalogues. Travel plans are stored in a memory of cases, which is obtained from ranking travel items ex- tracted from catalogues. MAPWEB: MULTI-AGENT PLANNING ON THE WEB This section describes in detail a B2C application called MAPWeb (Multi-Agent Planning on the Web) that is able to gather and reuse information from Web sources to integrate partial informa- tion into a common solution (Camacho, Aler, Borrajo, & Molina, 2005; Camacho, Borrajo, & Molina, 2001). MAPWeb Architecture $Q\ 0XOWLDJHQW 6\VWHP 0$6 GH¿QHG XVLQJ MAPWeb can be implemented using one or several teams. A specialized agent, CoachAgent (CCH), manages every team. To manage the different teams it is necessary to use a single ManagerAgent (MNG). This agent (known as Agent Name Server, ANS, i n ot her a rch itec tu re s) i s used to manage the insertion and deletion of other agents in/from the MAS. Therefore, in MAPWeb (as in other MAS frameworks), any system implemented will need at least the following agents to work properly: • Control Agents: manage the different agents in the system. There are two types: • ManagerAgent (MNG): This agent is similar to any ANS that can perform the following roles in the system: responsible for adding and removing other agents from the system; controls which agents are active in the agent society; groups agents in teams. When any agent requests to be inserted in the society, the MNG determines which teams require this agent. • CoachAgent (CCH): Controls a team of agents, guaranteeing stability and smooth operation of the active agents. These agents report problems to the MNG. When a new agent is required for the team, they guarantee that the 788 Applying Information Gathering Techniques in Business-to-Consumer and Web Scenarios yellow pages of the team members are coherent. • Execution Agents: These agents are respon- sible for achieving different goals within the system. To coordinate different teams of agents it is possible to include a new skill in the control module of the agents. Currently, there exist different kinds of execution agents, including those that are able to use a planner to solve problems (PlannerAgents), information agents that can retrieve Web data (WebAgents), agents that can interact with the RoboSoccer simulator, etc. Figure 2 shows the general architecture for any MAS in MAPWeb. The main characteristics of any MAS implemented in this way can be summarized as: • Agents in the system use message-passing to communicate with other agents. • All the agents have the same architecture and are specialized in different tasks through the implementation of different skills. • Although the communication language is the same for all the agents in MAPWeb, it is possible to distinguish two different types of communication messages. On one hand, there are control messages whose main goal is to manage the behaviour of the system (control communication in Figure 2). On the other hand, execution messages are used to share knowledge and tasks among the agents, to achieve desired goals (execution communication). To establish the MAS, it is necessary to perform the following steps: 1. First, the MNG is executed. 2. Agents in the system need to register them - selves with the MNG. Once a CCH has reg- istered, the MNG will select the necessary execution agents from its white pages and will build an operative team. If there are not enough agents, the CCH will wait for them. To build a team the MNG selects the execution agents and provides the necessary information to the CCH. Once the informa- tion on the agents has been stored in the CCH’s yellow pages, it updates the yellow pages of its execution agents. To select the necessary agents to build a group, the MNG uses the Ontology of the CCH agent. 3. Once a team is built, the execution agents can only communicate with the agents be- longing to its team or with its CCH. Figure 2. MAPWeb multi-agent architecture 789 Applying Information Gathering Techniques in Business-to-Consumer and Web Scenarios MAPWeb Application Domain Although the agent and multi-agent architecture proposed in MAPWeb has been successfully applied in other domains (such as robo soccer and genetic programming), the architecture was initially designed to gather and reuse information extracted from heterogeneous Web sources. The ¿UVWDSSOLFDWLRQGRPDLQRI0$3:HEZDVUHODWHG to the problem of planning user travel. The high number of companies connected to tourism activities, such as hotels, car rental, trans- SRUWÀLJKWVWUDLQEXVHVPXVHXPVWKHDWUHV etc., …, that have used the World Wide Web as a communication gateway between their resources and potential users, has changed the concept of WUDGLWLRQDO WRXULVP ,Q IDFW ³WUDYHO DQG WRXU- ism represents the leading application in B2C (business-to-consumer) e-commerce” (Fodor, Dell’Erba, Ricci, & Werthner, 2002, p. 1). The tourism domain is complex and heteroge- QHRXV,WLVSRVVLEOHWR¿QGKHWHURJHQHRXVLQIRU- mation such as maps, textual information (about places to visit), schedule and fare information about hotels, transports, etc., … It is possible for a n y u s e r t o c o n s u l t (a n d e ve n t o b u y) a n y t h i n g t h a t he or she could need in their travel experience. All of this information needs to be appropriately managed and integrated to provide useful informa- tion. Different systems have been implemented to deal with the tourism information in the Web, such as intelligent electronic tourist guides (e.g., GUIDE — Cheverst, Davies, Mitchell, Friday, & Efstratiou, 2000; cyberguide — Abowd, Atke- son, Hong, Long, Kooper, & Pinkerton, 1997), travel assistants (Ambite et al., 2002; Yang, Yang, Denecke, & Waibel, 1999), and context-aware systems for tourism (Zipf, 2002). Although other CI techniques such as Case- Based Reasoning or Information Retrieval have been successfully used in this domain, we have modelled the tourism domain as a planning do- main, where the user needs to create a plan to travel EHW ZHHQVHYHUDOSODFHV7 KLVPRGL ¿FDWLRQDOORZV us to use classical planning techniques, both to manage access to the Web companies that provide the information, and to look for heterogeneous solutions (integrating into a common solution the retrieved information from the Web). The main goal of this approach is to allow the integration of CI techniques, such as learning or planning, with Information Gathering techniques into a real and complex domain. Our MAS model uses a set of deliberative and cooperative software agents which integrate CI techniques and the Web to gather information (Camacho et al., 2005). The reasoning techniques used by those agents are planning (planning agents), and Case-Based Reasoning (Web agents and planning agents). The e-tourism domain has been modeled as a planning domain and has been included as part of the knowledge (ontol- ogy) shared by the agents. This domain uses a set of operators to work with the different topics in tourism such as hire rental car, book room, REWDLQÀLJKWWLFNHWVHWF8VLQJSUHYLRXVRSHUD- tors, specialized agents build skeletal plans that represents the different steps that may be executed by the user. The agent uses this skeletal solution to achieve several goals: • To coordinate other agents (user, planner, and Web agents). The planning agent uses the skeletal plan (abstract solution) as a template to decide what agent will be requested for help. • To share the information stored in different agents (old plans in other planning agents, or records retrieved by Web agents). • The use the skeletal plan as the integration VWUXFWXUHWREXLOGWKH¿QDOVROXWLRQWRWKH travel problem. The heterogeneous informa- tion provided by Web agents are integrated LQWR VSHFL¿F VROXWLRQV XVLQJ WKH VNHOHWDO plan. • To cooperate in the solving process with other planning agents. The planner agent can divide the initial problem into sub-problems 790 Applying Information Gathering Techniques in Business-to-Consumer and Web Scenarios with lower complexity (i.e., a problem with three independent goals can be divided into three one-goal problems), and request help from other planning agents. Figure 3 shows one possible MAPWeb topol- RJ\RUFRQ ¿JXUDWLRQZKHQWKHV\VWHPLVDSSOLHG in the tourism domain. Two operative teams managed by a MNGEXLOGWKLVFRQ¿JXUDWLRQDQG every team is locally managed by a CCH. Team_1 has the minimum set of agents to be operative, whereas Team_ 2 is built by K UserAgents, P PlannerAgents and J WebAgents. In addition, the following execution agents are needed: • UserAgents are the bridge between the us - ers and the system. They only implement basic input/output skills to acquire problem descriptions from users and to show the solutions found to them. • PlannerAgents are able to solve planning problems using the information gathered from the Web. • WebAgents are able to provide the requested Web information such as a set of relational records to the PlannerAgents using wrap- ping techniques. These agents use caching techniques to optimize the access to the Web. Although MAPWeb is a generic architecture that combines Web information retrieval with planning, its skills are better understood in a particular domain. In this section, we will use the e-tourism domain, where the goal is to assist a user in planning his or her trip. MAPWeb’s processes can be described as follows. First, the user interacts with the UserAgent to input their query. The query captures information such as the departure and return dates and cities, one way or return trip, maximum number of transfers, and some preference criteria. This information is sent to the PlannerAgent, which transforms it into a planning problem. This planning problem retains only those parts that are essential for the planning process — this is called the abstract representa- Figure 3. MAPWeb e-tourism topology 791 Applying Information Gathering Techniques in Business-to-Consumer and Web Scenarios tion of the user query. Then, the agent generates several abstract solutions for the user query. The planning steps in the abstract solutions need to be completed and validated with actual information t h a t i s r e t r i e v e d f r o m t h e We b. To a c c o m p l i s h t h i s , the PlannerAgent sends information queries to specialized WebAgents that return several records for every information query. Then, the Planner- Agent integrates and validates the solutions and returns the data to the UserAgent, which in turn displays it to the user. A detailed description about both the utilization of CI techniques, such as plan- ning or machine learning, used by PlannerAgents to reason with the retrieved information, and the coordination processes among different agents in MAPWeb through message passing can be found in Camacho, Borrajo, Molina and Aler (2001) and Camacho et al. (2005). Currently, MAPWeb is able to access, gather, and reuse information from the following Web sources: • Flight companies: Iberia Airlines, Avi- DQFD$LUOLQHV$PDGHXVÀLJKWV$LUOLQHV ÀLJKWV • Train companies: Renfe, BritRail, RailEu- rope. • Rental Car companies: Avis, Hertz, Ama- deus car rental, 4Airlines car rental. • Hotel companies: Amadeus hotels, 4Air- lines hotels. 7KHVH:HEVRXUFHVFDQEHFODVVL¿HGLQWRWZR main groups: Meta-search systems such as Ama- deus, 4Airlines, BritRail and RailEurope, which extract information from several companies, and individual sources which belong to a particular company (Iberia, Avianca, Renfe, etc.). Agents’ Characteristics The previous section has described the roles of control and execution agents used in MAPWeb, this section provides a description about the VSHFL¿FFKDUDFWHULVWLFVRIWKHVHDJHQWV7DEOH S U RY LGH V W K H V S H F L ¿F FK D U D F W H U L V W LFVRI W K H F RQW U RO  agents. Table 1 shows how it is only necessary to implement in the MNG agent simple acts such as insert or delete, because the main role of this agent is to manage the correct insertion and deletion of the agents into the system. This agent distributes the agents into operative teams (it tries to implement teams with at least one operative agent). However, the number and type of agents that implement a particular team could change in time. For instance, a particular WebAgent might not be operative for network problems (the Web server is down), so the CCH could suspend temporally the agent, and resume again when the problems has disappeared. The act: [Ask-for-agent] is used by the CCH agent to ask for the particular type of agent that is neces- sary for the correct work of the team. Finally, the Knowledge Base and the Yellow Pages of the agents stores all related information with the managed agents. The skills implemented in the MNG agent al- lows it to modify (in number and type) any team of agents. This agent uses a policy to distribute new agents and to build the teams, or to share a particular agent in several teams (by providing its direction to the CoachAgents of the teams). The skills implemented for the CCH agents allow those agents to manage a team, and to request help from the MNG if any agent (PlannerAgent, WebAgent, UserAgent) is necessary for the team to work correctly. Table 2 provides a description about the char- acteristics of the execution agents in MAPWeb. As Table 2 shows, the implemented UserAgent is the most simple execution agent, it has only a set of graphical interfaces to allow communica- tion between the user and the system. The most remarkable characteristic in these agents is the implementation of a fuzzy skill that is used like DFODVVL¿HUWRVHOHFWDQGRUGHUWKHPRVWSURPLVLQJ solutions found (this algorithm uses some char- DFWHULVWLFVSURYLGHGE\WKHXVHUVLQWKHLUSUR¿OHV 792 Applying Information Gathering Techniques in Business-to-Consumer and Web Scenarios 7DEOH6SHFL¿FDWWULEXWHVIRUFRQWURODJHQWV Skills Knowledge Base Communication Module ManagerAgent (MNG) - Insert new agent - Delete agent - Team management policy Stores the information about all the agents in the system, naming information, skills, teams, etc… Is implemented using the Kqml-ACL language and TCP/IP protocol CoachAgent (CCH) - Insert new agent in the team - Delete agent in the team - Suspend a particular agent - Resume a particular agent - Ask for new agent Their KB stores the specific information about their execution agents and the contact address of the MNG Kqml-ACL language and TCP/IP protocol 7DEOH6SHFL¿FDWWULEXWHVIRUH[HFXWLRQDJHQWV Skills Knowledge Base Communication Module WebAgent - Automatic-Web-access. It allows access, retrieval and translation into an relational format the data found in the Web sources. - A caching technique. It has been implemented using a database of retrieved records (only stored those records that are finally sent to the PlannerAgents) This module is built with the context information necessary to access and manage their related Web sources, and by a Case-Base of records retrieved from these sources Kqml-ACL language and TCP/IP protocol PlannerAgent - Case-Based Planning. It allows storage, retrieval and retention of old successful plans that can be used later by this agent (or by other PlannerAgents in the team) - Planning. Permit the use of planners (we actually use Prodigy 4.0) as main reasoning module - Cooperation. Allows to request help in the solving process to other PlannerAgents The information to translate the user problem into an standard representation, the definition of the planning domain (e- tourism), and the Plan Base is stored, and managed, by these agents Kqml-ACL language and TCP/IP protocol UserAgent - User/system communication. It allows to provide the problems, user characteristics (profile), and to show the solutions found - Fuzzy classification. A fuzzy algorithm is used to sort appropriately the solutions found by the system Store information about the users (they can adapt the behaviour of the system defining his/her profiles), and information about the PlannerAgents that can help to solve a problem Kqml-ACL language and TCP/IP protocol 793 Applying Information Gathering Techniques in Business-to-Consumer and Web Scenarios RUDSUHGH¿QHGSDWWHUQZKLFKXVHVFKDUDFWHULVWLFV such time of travel or cost). The main goal of the PlannerAgents is to reason and deal with the Web information gathered by the WebAgents and build solutions for the given problems. For this reason, several techniques such as Planning and Case-Base Planning have been implemented. The cooperation among PlannerAgents is also allowed. Finally, the main characteristics of the WebAgents are the Automatic-Web-access and the Caching technique implemented. The automatic Web access is implemented through a set of specialized Wrappers that allows those agent to access, retrieve and store in a standard format the information stored in the Web sources. In addition, the implementation of a Case Base allows those agents to minimize the number of times they need to access the Web. FROM MAPWEB TO MAPWESC 7KLVVHFWLRQGHVFULEHVKRZRXUVSHFL¿F%&,Q- formation gathering application (MAPWeb) can be migrated into a new Web Services oriented architecture called MAPWeSC (Multi-Agent Planning through Web Service Composition) to adapt an old agent-based Information Gathering application into current Web technologies. Problems and Pitfalls in MAPWeb When any Web Information Gathering MAS system, such as MAPWeb is deployed several problems (related to their domain application) need to be dealt with. Some of those problems FDQEHEULHÀ\VXPPDUL]HGDVIROORZV • There exist several problems related to important characteristics such as, proac- tiveness, security or robustness, in those Multi-agent systems. Although several CI techniques (such as Machine Learning or planning) are used to improve on previous characteristics it is necessary to study (using the current Web technologies) how to ask the following questions: • What happens when several Web sources are temporally down? • How could any agent be proactive in the Web? • There are important problems related to the management of the information retrieved from the Web sources. Two main problems need to be solved - how to extract the in- formation from several Web sites, and how represent this information in a coherent way. Therefore, for any IG Web systems it will be necessary: • To apply several information extrac - tion techniques that allow the agents to automatically gather information from a particular Web site. • To represent (using a standard format) the information gathered from the Web, to allow this information to be shared between different information agents in the system. Although previous problems have been tradi- tionally solved using techniques such as Wrapper technologies or Machine learning (to automati- cally extract the information), ontologies (to repre- sent the information), etc., the implementation of new Web technologies (such as XML, SOAP, Web services) provide a new frameworks that could be used to design new solutions to old problems. The previous problems could be solved using Web services technologies as follows: • Robustness and security characteristics can be improved using the Web services secu- rity features that provide message integrity, DXWKHQWLFDWLRQDQGFRQ¿GHQWLDOLW\PHVVDJH session security, etc. • Proactiveness in a multi-agent system can be improved using the facilities of Web services discovery. The Universal Descrip- . problem into an standard representation, the definition of the planning domain (e- tourism), and the Plan Base is stored, and managed, by these agents Kqml-ACL language and TCP/IP protocol. initiative. An agent is capable of handling complex, high-level tasks. The decision as to how such tasks are best split up into smaller sub-tasks, and in which order and way the sub-tasks are best. commu- nicate amongst themselves and coordinate their activities. Coordination and communication are FHQWUDOWR0$6 IRUZLWKRXWLWDQEHQH¿WVRI interaction vanish and a group of agents quickly