Agentbased BROkerage Services in  Electronic commerce

Objectives of the Research Project

Electronic commerce  exchange of products over the Internet  consists of brokerage and  negotiation phases.It has been developed a lot because of  the facilities available on the Internet where the amount of products always increases. Consequently, a user needs a tool to guide him searching for the right partner.

The ABROSE system is used to find the right "interlocuteur". It helps  the brokerage offering a customer, a list of relevant content providers. It also facilitates the sending of  targeted publicity for new content providers' offers.

The brokerage function of ABROSE is a software designed in accordance with the AMAS theory. The architecture of ABROSE consists in a set of several adaptive multi-agents systems. All users (customer or content provider) is represented by an agent. This agents has an incremental management of the  profile about the user itself but also about others agents. This profile is automatically updated during transactions realized by the user. The originality of ABROSE is in the automatic management of the user's profile and the use of the experiences of all the transactions to improve the mediation function.

Context of the ABROSE Project

An electronic market place consists of an important set of content providers and customers which have dynamic offers and needs. Abrose (Agent based BROkerage SErvices in electronic commerce) is an electronic commerce tool for mediation of services which are based on agents.

The main idea is that a collective memory is elaborated in a population of content providers and customers represented by agents. This collective memory consists of the individual experiments or recommandations formuled by users. Abrose tries to model and to treat this collective memory (and also the individual memories) in an electronic market place to facilitate the quality of the exchanges. The main functionalities of Abrose are :

• For customers simplified interactions, an aid assistant, a spontaneous notification of new offers, a help for surfing and for request formulation, and finally the proposition of relevant content providers' list answering  the request made by the customer. 
• For the content providers, a targeted publicity of offers towards only the interested customers, an amount of information about the customers'real interests and about the offers of the market.

System Architecture

To reach its aims, an architecture with several levels of multi-agent systems was specified and realised. the platform Abrose V2.0 is implemented with Java  JWS1.1.3, The communication between the multi-agent system and the rest of the system is developped with OrbixWeb3.199.

The brokerage software which includes the multi-agent systems, runs under Solaris 2.6. The customer is a  Personal Computer with the standard navigator Netscape 99.

The brokerage software consists of three levels of self-organisator multi-agent systems: 

1. The upper level of Abrose is a multi-agent system related to the sites (Mediation agents) which communicate by message sending.
2. All the sites are  multi-agent systems of intermediary level (Transaction agents) consist in agents which interact in them. All of these agents represent either a customer or a content provider.
3. To allow relevant interactions in each of these two levels, the agents reason with private and dynamic knowledge they have about each others : the beliefs. The beliefs are managed in a multi-agent system called belief network which is private to each transaction agent. All these belief networks are at the lowest level. The cooperative self-organisation approach used to implement these multi-agent systems, allows an optimal, individual and collective satisfaction although a global knowledge of the system does never exist.

Scenarios of Demonstration

ABROSE is a multi-agent system where each agent cooperates in using its own belief network, to identify the offers which can be interesting for customers, or requests which can be interesting for content providers. The requests and the offers are expressed in a free text and the results provided are ordered in accordance with the relevance they have in the belief network.. 
The scenarios of demonstration below show the  functionalities of Abrose, at the service quality level, proposed to the users and its application in two domains : electronic commerce "business to business" and teleworking. For the demonstration 20 content providers and 50 customers are on the market place.

Scénario 1 : "Need a Translation Service"

In this scenario, the functionalities of Abrose showed focus on the search for relevant content providers related to a given request and the facility to formulate a request for a customer. The user connects himself to the Abrose system. He selects the menu in respect of the request insertion and fills up the fields subject and text of the request.
Abrose analyses its beliefs to know if there exists one or more content providers relevant to the answer of the request. The content providers which can answer, are presented to the user, ordered according to their relevance. The user chooses the content provider which seems the most relevant. This selection implies the automatic connection to  the content provider 's site and interface. The user can after be invited to complete his request and to do a transaction : selling of the product or of the service. During the transaction, the user 's actions are registered and used by Abrose to learn and to update its belief about the content providers and the customers when the users evaluates his transaction.

An example of this kind of scenario is realised with a user representing Plantagenet Limited Ltd (UK). This compagny is specialized in new generation of computers. Plantagenet searches someone to translate in Swedish a technical document written in English. The steps are the following :

• Plantagenet user connects itself to ABROSE system.
• The users selects the menu "Request/Offer Registration" to write his request with the following subject: "Translation" and the following  text: "I need for Translation of Technical Documents on Supercomputers (about 50 pages long) from English into Swedish". 
• Abrose gives the relevant content providers' list : "Web+Translation, Transcription, Written+Translation, Interpreting". 
• After the user's choice, the connection to Tradezone system is done and the transaction is realised. 
• Finally, in order to facilitate the system to learn, the transaction is evaluated to "good"

To show the learning at the belief network level, (the belief network is inside the user's transaction agent), this request is given one more time. The content provider's name choosen is different and the evaluation of the transaction is now "bad".

The evolution of the transation agent belief (represented Plantagenet) between the first and the second transaction, could be seen  doingthe same request one more time.

See also the section devoted to "Documents" (demonstrations)

Scenario 2 : "New Offers"

In an offer scenario, the content provider has something to sell and wants to use the Abrose functionalities which allow him to reach possible customers. As in the previous scenario, the user connects itself to Abrose and gives a subject and a description of his offer. Abrose analyses the belief network of the mediation agent in order to identify interested customers. These customers are displayed for the content provider, who chooses to which customers he wants to send ths offer. When a selected customer will connect to Abrose, the system will announce to him, he has received an offer. After he will evaluate the relevance of the received offer.

To show an example of an offer propagation, three requests are realised, for which Abrose cannot propose content provider.
there are three users: Glize and Barrett ask:"I search for senior programming to develop a Web site" and Einsiedler asks: "I search for a development of C++ software". After, Yamina a specialist of Web site development, who searches for a job, writes an offer to Abrose via the content provider Programming. 

The scenario is like the following :

• The content provider Programming connects to ABROSE 
• He proposes the Yamina's offer with the subject: "Web development" and the text of the offer : "I am looking for some Web development work to start mid-may, Yamina , at a senior programming level". 
• Abrose gives  her only two possible customers: Glize et Barrett. 
• Programming chooses to send the offer to the two customers

Some time after, when the customersGlize or Barrett connect toABROSE, they can evaluate the  relevance of Programming offer in the menu "Visualisation Request/Offer. In this scenario, the learning at the mediation agent level is shown because it stores the unsatisfied resquests of the transaction agents Glize, Barrett and Einsiedler.

The originality of Abrose and its main advantages are :

• the system does not use a predefined ontology, the ontology builds itself by the learning of the system
• the system is not dedicated to a selling  a type of product. it could be used for any domain
• the system is not dedicated to any language, the contents providers can describe their services in the language of their choice. The request should be realised in the same language
• the system is open, the creating or removing of transaction agents is dynamically realized. the discover of a new agent or the removing of an agent is learned by other agents during time.
• the system adapts  itself constantly to the evolution of the market place.
• each agent has a partial knowledge of the others.

References about the Project

Conference Papers

• GLEIZES Marie-Pierre, GLIZE Pierre - ABROSE: Multi Agent Systems for Adaptive Brokerage - Fourth International Bi-Conference Workshop on Agent-Oriented Information Systems (AOIS-2002), 27-28 May 2002, Toronto (Ontario, Canada) at CAiSE'02.
• GLEIZES Marie-Pierre, GLIZE Pierre, LINK-PEZET Jo - An Adaptive Multi-Agent Tool For Electronic Commerce - The workshop on Knowledge Media Networking IEEE Ninth International Workshops on Enabling Technologies : Infrastructure for Collaborative Enterprises (WET ICE 2000)  14-16 June 2000 Gaithersburg, Maryland
• GLEIZES Marie-Pierre, GLIZE Pierre - ABROSE : Des systèmes multi-agents pour le courtage adaptatif - 8ièmes Journées Francophones d'Intelligence Artificielle Distribuée et des Systèmes Multi-Agents, Saint-Jean de la Vêtre France 2-4 octobre 2000
• ATHANASSIOU Eleutherios, LEGER Alain, GLEIZES Marie-Pierre, GLIZE Pierre - Abrose : Adaptive Brokerage based on Self-Organisation Services and Users - Poster MAAMAW'99
• ATHANASSIOU E., CHIRICHESCU D., GLEIZES M.P., GLIZE P., LAKOUMENTAS N., SCLENKER H., LEGER A., MORENO J.I. - Abrose : A Co-operative Multi-Agent Based Framework for Marketplace - IATA, Stockholm, Sweden August 1999
• CARRE Jérôme, MACHONIN André, GLIZE Pierre - Un système multi-agent auto-organisateur pour l'apprentissage d'un profil utilisateur - 7ièmes Journées Francophones d'Intelligence Artificielle Distribuée et des Systèmes Multi-Agents, Saint-Gilles La Réunion, France, 8-10 novembre 1999
• CARRÉ Jérôme, GLEIZES Marie-Pierre, GLIZE Pierre - A user profile learning mechanism from users profile - Proceedings on European Conference on Cognitive Science, Siena, 1999
• EINSIEDLER H.J., GLEIZES M-P., GLIZE P. - Abrose : A Co-operative Multi-Agent Based Framework for Electronic Marketplace - Infowin-Infobridge: Book about "Agent Technology", ACTS related publication November 1999
• GLEIZES M.P., LEGER A., ATHANASSIOU E., GLIZE P. - Abrose : Self-Organization and Learning in Multi-Agent Based Brokerage Services - 6th International Conference on Intelligence and Services in Networks, IS&N'99, Barcelona, Spain. Proceedings pp 41-54., Lecture Notes in Computer Science 1597, Springer 26-28, April 1999
• GLIZE Pierre, GLEIZES Marie-Pierre, LÉGER Alain - Brokerage communication in a cooperative multi-agent based mediation service : one example in Abrose - Foundation for Intelligent Physical Agent - Proceedings CFP6_016, 1999
• LEGER A., GLIZE P., GLEIZES M.P., EINSIEDLER H.J. - Abrose : Multi-Agent Based Brokerage Services in Electronic Commerce - Agent Link, Barcelona, Spain, 20 September 1999
• CAMPS V., GLIZE P., GLEIZES M.P., LEGER A., ATHANASSIOU E., LAKOUMENTAS N. - A Framework for Agent Based Information Brokerage Services in Electronic Commerce - EMMSEC, Bordeaux (F), 28-30 September 1998
• LEGER A., GLIZE P., GLEIZES M.P., CAMPS V. - ACTS-Abrose and FIPA - Preliminary studies - FIPA, Osaka, Japan 15-19 April 1998

Dissemination

• EINSIEDLER Hans, GLEIZES Marie-Pierre, LEGER Alain, LOPEZ Jorge - 7th International Intelligence in Services and Networks Conference, IS&N 2000, February 23-25, 2000, Athens, Greece
• GLEIZES Marie-Pierre, GLIZE Pierre - Abrose : Des agents personnalisés adaptatifs pour le commerce électronique - Forum France Telecom « Internet et mobilité » - Mai 2000
• GLEIZES Marie-Pierre, GLIZE Pierre - Profil adaptatif d'un utilisateur par système multi-agent pour le commerce électronique, Agent-based BRokerage Services in Electronic commerce : ABROSE - Journée Industriels IRIT, Toulouse, Septembre 2000
• GLEIZES Marie-Pierre, GLIZE Pierre, MACHONIN André - ABROSE : Un outil de courtage adaptatif supporté par un système multi-agent - Démonstrations pour les 8ièmes Journées Francophones d'Intelligence Artificielle Distribuée et des Systèmes Multi-Agents, Saint-Jean de la Vêtre, France, 2-4 octobre 2000