MARI: Multi-Attribute Resouce Intermediary

( Gaurav Tewari, Pattie Maes [MIT Media Labs])



 

ABSTRACT

        MARI ( Multi Attribute Resource Intermediary ) is proposed to improve online marketplaces , specifically those involve buying and selling of non-tangible goods and  services. It  is an agent based intermediary architecture intended as a generalized platform for the specification and brokering of heterogeneous goods and services. It basically comes under three areas
                                    - multi- agent system
                                    - highly mediated communications
                                    - electronic commerce .
        But it is primarily positioned from a "highly mediated communications" perspective . MARI builds upon  multi-attribute utility theory formulations to model relative user preferences and quantify tradeoffs. Here the user can specify the range of attributes he can use and also the Utility Function for each attribute. Due to this a buyer is made to show his hidden preferences through his preferences. For example, if user's thinks that he wants to get a product for a perticular price only, then he will give almost equal values for maximum price and minimum price, which inturn used by the MARI to match it to a seller whose selling price is almost same as buyer's wish.
 
 



 

NEED FOR MARI

         Unlike most online shoppeing systems which generally operates in only one stage of the online process it operates in three core stages of online shopping
                                    - product brokering
                                    - merchant brokering
                                    - negotiation

        MARI not only relates to most of the classical models  it also improves them As follows :

        It relates to first generation price-comparison system but does more than the rudimentary functionality afforded by such tools . It goes beyond just bid and ask prices to include the attributes of transaction parties as dimension for consideration and differentiation .

        MARI relates to second generation value comparison shopping systems where it offers advanced decision suport engine based upon multi-attribute utility theory , that meaningfully faciliatetes the exchange of comples and heterogeneous products.  It improves upon this model  by

                                1. allow both parties (buyers and sellers) to search for an optimal transaction partner

                                2. Automates the match making between buyers and sellers .

        Also it supports a non-linear and iterative user interaction model , that accurately reflects the true nature of real-life transactions.

        It relates to online negotation systems and auctions , such as Kasbah, and differs from them in proposing an integrative negotiation protocol and interaction model.

        MARI relates to dynamic pricing of inventorie, where sellers  dynamically  shift their  valuations when demand is price sensitive  and stochastic  and seller's objective is to maximize expected revenues. It is based on  flow algorithms encountered in combinatorial optimization and network theory. Also uses market-oriented allocation mechanisms and economic theory in general and game theory in particular . To formulate MARI problem  in economics terminology with optimization heuristics , such as maximization of aggregate surplus that drives directly from the literature .

        MARI is idle model when negotiations will be hightly complex and participants will engage in integrative negotiation over various aspects of a transaction price being only of the many considerations .

        In MARI each seller has the ability to differentiate her product or services from those of other sellers . The market structure is monopolistic in the sense that each seller has ability to set her/his  own price . On the other hand , each seller still competes in terms of price and range of product offered , with other sellers since they are all effectively trying to find transaction partners from a common underlying set of buyers . Moreover there are no barriers to entry, and new sellers are free to enter the market. In this way, the market structure also resembles that of a competitive industry.
 
 



 
 

 CRITICAL ASPECTS OF MARI

                    - MARI trades non-tangible goods and single item

                - Type of bid it uses is sealed bid

                - There are multiple rounds of negotiations in MARI

                - Revenue of any one ( buyer, seller, market) can be maximized

                    (In our implementation we have maximized Market Revenue, which in turn maximizes Revenues of buyers and sellers together)

                - Optimization problem has been handled using Matching Algorithm


MAJOR FUNCTIONAL COMPONENTS OF MARI

        Attend : User Interface manager is handles all intractions between user and MARI. It presents the appropriate interface to the user , such that the system is able to adequately capture all relevant parameters .

        dataBuyer, dataSeller : Each buyer or seller is represented within MARI by an agent . Each agents is customized to the needs and desires of its owner , and attempts to advocate on the owner's behalf when finding suitable transaction partners. These agents are then used by the system to coordinate the preferences and intrests of each party involved .

        MariServer : It creates an instantiation of MariAlgo specific to their product domain . When instantiating MariAlgo, the market maker specifies the product ontology as well as set of parameters which determine how user utility functions are modeled and what heuristics are used in match process .

        MariAlgo : It finds the optimal pairing such that Market Revenue is maximized. For that to happen number of matchings will be maximized, which inturn maximizes the revenue of both buyer and seller.
 



 
 

HOW MARI WORKS ?

            Step 1: Buyer/Seller specifies all of his preferences to MariServer using interface provided

            Step 2: MariServer collects all those requests and stores them.

            Step 3: After every time T it instantiates MariAlgo for each product

            Step 4: Then it collects result and updates the corresponding entries
 

Details of MariAlgo:

            Input: List of buyerAgent ds and List of sellerAgent db

            Step 1: For each buyer find the hidden weights for each attribute

                         Relative Weight(i) =1- (Permissible Range of attribute i/Possible range of attribute i)
                         Weight(i) = Relative Weight(i)/( Sum Relative Weight(j) for all j)

            Step 2: A table A is created such that

                        A[i][j] = incompatibility( buyer i, seller j)
                        incompatibility(i,j) = Sum of [  cost of each attribute of j, using referential function of i ]

            Step 3: Find optimal pairings such that total cost is minimized

            Step 4: Return optimal pairings
 

The optimization problem:

                   Minimize  Sum [ CijXij ]
                   Subject to

                                Sum[Xij] = 1 for all i
                                Sum[Xij] = 1 for all j

                Xij = {0,1}


RELATED WORK

Kasbah:
           Authors: A Chevez, P  Maes

            This paper proposes a E-Market system which can improve online trading to a large extent.  In this users can create autonomous agents which will do the trading for them. This approach is more natural compared to other online trading systems.

            WHenever a User creates a agent(buying/selling) he enters some parameters like BestPrice, WorstPrice, Time, NegotiationStrategy.  Once created, the agent will look for potential Traders (whose list will be provided by the server) and starts negotiating with them. While negotiating it uses the strategy provided by the User.

            NOTE: As a part of Programming Assignment we made the following improvements:
                    1. Agent1 will stop when they get a fellow Agent2 who meets its Price. But in our implementation if the number of Agents who meets CurrentPrice is more than some Threshold, Agent1 will "Improve" it's CurrentPrice.
                    2.  When  NegotiationStrategy is Quadratic or Cubic, the rate of change will only depend on time. But in our implementation rate will change according to    | No of times CurrentPrice has been Improved- No of times it has been Worsened|
 
 
 

Agents that Buy and Sell:Transforming Commerce as We Know It

             Authors: Pattie Maes, Robert H Guttman and Alexandros G Moukas

           Agents that automate the work of humans have been used in many fields to decrease the overload on humans. Recent Usage of these Agents in Ecommerce has revolutionised the way the transactions happen in the E-market place. Here a survey of various buying and selling agent technologies and e-commerce systems that use these automated agents in a general buying model is taken up.

         With the evolution of Internet and World Wide Web the number of transactions on the Internet have increased phenominally. But due to human factor involved in the transactions the throughput is less. So with the automation of some tasks which now involve human interaction, the cost of transactions can be decreased. Software agent technologies differ from traditional software by being personalized, semi-autonomous, continuously running. These characteristics optimize the whole buying experience and revolutionize e-commerce. These automatically collect required information about products and vendors which satisfy the given constraints, evaluate the offers, take decisions about the products and vendors, negotiate the transaction, make orders and finally make payments.

        The agents have a major role as Mediators in Ecommerce under the Common Buying Behaviour model. Even though the approach is limited it offers an insight in the role to be played by these agents. There are many Buying Behaviour models but all of them contain six main activities. These activities and the role of agents played in these are described below:
Need Identification:-
        This Stage involves finding of products that satisfy the buyers requirements and he being informed on their availability. Here agents can be useful in requirements that are repetitive or predictable. These agents are like monitors that run continuously and monitor a set of activities and take actions to activities that satisfy a certain pre-requisite. Examples include Amazon.com,Stock Market and Ecommerce sites, which inform potential buyers about the arrival of new product which has been specified the customers.
Product Brokering:-
        This Stage involves evaluating different product offers and making a decision on what to buy given a set of constraints by the user. Here agents can be given a set of constraints and they take the optimal decision. The agents can be divided into mainly constraint satisfaction engines or collaborative filtering techniques. The constraint satisfaction engines search through a large feature space and find the set of offers that satisfy a set of hard constraints and in the order of satisfaction of soft constraints. The collaborative filtering techniques work through collaboration of different agents and filtering out the useful offers. Other techniques involve Data Mining and Rule Based Systems which make patterns in customer behaviour and personalize the buying experience.
Merchant Brokering:-
        This Stage involves using the set in the before stage and gauging the merchant information to find the best merchant for the offer. The Agents can be made to find an optimal alternative out of the possible merchant alternatives. The limitations of agents here is that they donot take into account various value added services in account and possibility of vendors detecting an agent and blocking its requests. These limitations can be avoided by using wrappers and other techniques.
Negotiation:-
        This Stage involves deciding the terms of the transaction. The negotiations vary in duration and complexity of the market. The need to automate this stage comes due to frustration and time consuming nature of negotiations and the costs incured. The Agents can negotiate to any length of time and they can be set to optimize certain terms of the transaction. These agents can be personalized to each transaction and to each customer. These agents can be even set to negotiate even value added services along with price and different additions can be made to make the decision process better like combining both product and merchant information to select the best product and merchant at the same time.
Purchase and Delivery:-
        Once the terms of the transaction are agreed, the purchase and delivery is taken up. The mode of purchase and delivery might influence the product and merchant brokering. The agents can be made to make offers depending upon the needs of the customer.
Product Service and Evaluation:-
        This Stage involves product service, customer service and evaluation of overall buying experience and decision. The agents can be used to do analysis in post-purchase phase to evaluate the overall experience of both the customer and seller.
        Basically these stages are an approximation of behaviour of real world system. Thus can be seen that software agents provide a way of decreasing the workload and make complex decisions. The Agents have reduced transaction costs and have revolutionized certain areas of business. In future the agents may interact and make decisions on dynamic behaviour to make complex offers, thus revolutionising the whole way transactions are done on the Internet and World Wide Web.
 
 
 


CONCLUSION

        MARI and KASBAH are a good steps in the direction of a very effective, global, frictionless, completely automated marketplaces. But still a lot of work has to be done in the following areas:

            * Adding more intelligence in Agents
            * Automating all the processes in a transaction
            * Better payment systems
            * Combinotorial Trading, where a agent can sell/buy more than one item
            * Better algorithms to check monopoly
 
        A dream Marketplace would look like this:
 
            * A generic markeplace which just acts as communicator
            * There should be a lot of options ( better to say, real lifelike options) while creating a Agent
            * A capable end user should be allowed to create his own agents ( with his own negotiation algorithm )






BIBLIOGRAPHY

[1] Design and implementation of an Agent-Based Intermediary Infrastructure for Electronic Commerce. ( G Tiwari, P Maes)
[2] Kasbah: An Agent Marketplace for Buying and Selling Goods ( A Chavez, P Maes)
[3] Adaptation in Natural and Artificial Systems  (Holland)
[4] Agents that Buy and Sell:Transforming Commerce as We Know It ( Pattie Maes, Robert H Guttman and Alexandros G Moukas)