MultiAgent Systems Lecture 12 Computer Science WPI Spring

Multi-Agent Systems Lecture 12 Computer Science WPI Spring 2002 Adina Magda Florea adina@wpi. edu

Agents for e. Commerce Agents for information retrieval Lecture outline 1 Agents for e. Commerce 1. 1 e. Commerce 1. 2 Consumer's buying behavior 1. 3 Agents as mediators in e. Commerce 1. 4 Information economy 2 Agents for information retrieval 2. 1 Information agents 2. 2 RETSINA 2. 3 Web. Mate

1 Agents for e. Commerce 1. 1 Electronic commerce Components n interactive business and financial transaction n electronic cataloguing n electronic order tracking services n automatic billing and payment services n electronic funds transfer n vendor registration and electronic "brand naming" n automatic ordering, contracting and procurement n data mining of consumer information for customer profiling n advertising of products and customization of advertisements 3

n Transactions business-to-business-to-consumer-to-consumer Difficulties of e. Commerce The Web has a number of features that limits use as an "information market" Problems related to using the Web for e. Commerce: Trust Privacy and security Billing Reliability 4

1. 2 Consumer's buying behavior Marketing Consumer's Buying Behavior (CBB) research - a number of models of the consumer's behavior Most common stages; a simplification; some stages may overlap CBB - Guttman e. a. , 1998 Need investigation Product brokering Merchant brokering Negotiation Purchase and delivery Product service and evaluation 5

1. 3 Agents as mediators in e. Commerce Most appropriate for mediating behaviors involving information filtering and retrieval, personalized evaluation, complex coordination and negotiation Persona Bargain Auction Fish Logic Firefly Finder Jango Kasbah Bot T@T Market Need identification Product brokering Merchant brokering Negotiation Purchase and delivery Product service 6

(a) Comparison shopping agents Search online shops to find products, merchants and best deals Persona Logic n n n Product brokering guides the consumers through a large product feature space allows shoppers to specify constraints on a product and scores the products CSP engine: hard constraints and soft constraints 7

Firefly n n helps consumers find products uses "word of mouth" recommendations ACF = Automated Collaborative Filtering identifies the shopper's "nearest neighbours" and offers products highly rated by them Bargain. Finder n n n Merchant brokering the first agent for price comparison given a specific product, the agent requests its price from each of nine different merchant Web sites using the same http request as a Web browser Problem: some merchants block access to their prices; other merchants volunteer their prices 8

Jango n n n n helps users decide what to buy finds specifications and product reviews makes recommendations to the user performs comparison shopping for the best buy monitors "what's new" lists, watches for special offers Problem = Web pages are different; exploits: Navigation regularities Corporate regularities Vertical separation has 2 key components: a component to learn vendor description a comparison shopping component Solves the merchant blocking issue by having the product requests originating from each consumer's Web browser instead of a centralised site as in Bargain. Finder appear as requests from real customers 9

Product brokering and merchant brokering agents use information filtering techniques n n n content-based filtering, e. g. associative networks of keywords as in Jango constraint-based filtering, like in Persona. Logic, T@T collaborative-based filtering, like in Firefly 10

(b) Auction bots Agents that can organize and/or participate in online auctions for goods Kasbah Aim = develop a Web-based system in which users can create their own agents to buy and sell goods on their behalf User options: Create a new buying agent Create a new selling agent See currently active agents Create a new finding agent Browse the marketplace for active agents 11

n n Selling agent parameters set by the user: - desired date to sell the good - desired price to sell the good - minimum price to sell at - "decay" function of the price over time to determine the current offer price • anxious - linear function • cool headed - quadratic function • frugal - exponential function Buying agent parameters set by the user - date to buy the item by - desired price - maximum price - "growth" function of price over time 12

n n n Kasbah agents operate in a marketplace The marketplace manages a number of ongoing auctions matching requests for goods with offers Negotiation protocol - buying agents offer bids to sellers - selling agents respond with yes or no Tête-à-tête n n User agents negotiate across multiple attributes of a transaction, e. g. , warranty length and options, shipping time and cost, service contract, return policy, quantity, accessories, credit options, payment options Agents quantify those aspects using a multi-attribute utility function 13

Fishmarket n A virtual institutions corresponding to a traditional fish market which exists in Blanes (Girona) a small fishermen's village in Spain BA SA Buyer's register Goods show and auction Auct BM Credits and goods delivery 5 basic scenes BA = buyer's admitter SA = seller's admitter BM =buyer's manager SM = seller's manager Auct = auctioner SM Sellers' settlements 14

t Market operation (simplified) 1. Open auction and register sellers (SA) 2. Collect products from sellers (SM) 3. Collect buyers (BA) 4. Present products at price w (4. . 7 - Auct) 5. if silence then decrease w go to 4 6. if first bid w' w then adjudicate product 8. Verify credit (BM) go to 8 9. if not solvable (BM) 7. if two equal bids then fine or expell then increase w to x * w' go to 4 10. else sell product update buyer's credit (BM) update seller's credit (SM) 15

The first valid offer is the one to win the round An offer is valid if the bidder has enough credit to pay for that bid Fishmarket was also tested for closed bid auctions and Vickrey auctions Does not automate negotiation Problems with auction bots n n n Main difficulty - trust if: the agent really understands what the user wants the agent is not going to be exploited by other agent the agent does not end up with a poor agreement 16

1. 4 Information economy n University of Michigan Digital Library (UMDL) is structured as a collection of agents that can buy and sell services from each other n Treating a library as an information economy provides a framework for making decentralised decisions about allocation of limited information goods and services available n The services and protocols offered by UMDL infrastructure are called SMS = Service Market Society 17

The Service Market Society implements a multiagent information economy where agents buy and sell services from each other. Find phase Bid phase Query phase SCA 4 1 Label Query 5 8 QPA 6 Auction Match me with a seller at a price Registry Info resources AMA 3 UIA 2 Match me with a buyer at a price 2 6 QPA 7 CIA 9 Query 18

n n n Ontology of services SCA classifies the service description into a subsumption-based taxonomy SCA matches requests for services to "semantically close" descriptions Auction specification o type of good o timing requirements o terms - per-query or subscription (how is bundled) - topic, audience - redistribute or read-only (terms) - individual or library or group (to whom is sold) o how often the auction is cleared o price determination rule o what info is publically available 19

n QPAs bid their marginal cost = what it would cost them to provide another unit of the product Cost(query) = A * load 2 + B * load Marginal. Cost(query) = 2 * A * load + B n n n The Auction matches current lowest price seller with a buyer if the buyer's bid is above that price Once a transaction occurs, both buyers and sellers are removed from the active list and the QPA recomputes its marginal cost based on having an additional query to process Then QPA submits a new, higher sell offer to the auction 20

Strategic agents in UMDL n the bidding strategy is based on stochastic modelling n the model captures factors that influence the expected utility for the agent using Markov chains n a seller is likely to raise its offer price when there are many buyers or when it expects many buyers to come Learning agents in UMDL n Agents use past experience and evolved models of other agents to better sell and buy goods 3 types of agents: 0 -level agents, 1 -level agents, 2 -level agents - see Lecture #8 21

2 Agents for information retrieval 2. 1 Information agents Information agent = an agent that has access to at least one and potentially many information sources, and is able to collate and manipulate information obtained from these sources in order to answer queries posed by the user or other information agents Need of information agents n information overload n people get bored or confused by the Web n using Web is more for browsing than for reading 22

Several types of information agents Personal agents • provide "intelligent" and user-friendly interfaces • detect the user difficulties and help the user to get round the • • problems observe the user and learn his profile (preferences and habits) sort, classify and adminstrate e-mails, organis, schedule, plan user's tasks in general, agents that automate the routine tasks of the users Web agents • Tour guides • Indexing agents • FAQ finders • Expertise finders Search engines - human indexing - spider indexing 23

Cooperative information retrieval systems n n o o n Use information retrieval theory and AI Make information resources available by wrapping them with agents capabilities Every agent is expert with its own repository Agents communicate using an ACL Info. Sleuth RETSINA - mainly an architecture Middle agents or brokers Each agent advertise its capabilities to some broker Information brokers: matchmakers or yellowpages agents 24

2. 2 RETSINA 2. 3 Web. Mate Jinhui Jian's presentation 25

References n n n M. Wooldrige. An Introduction to Multi. Agent Systems, John Wiley&Sons, 2002, Ch. 11, p. 243 -266. R. Guttman, A. Mokas, P. Maes. Agents as mediators in electronic commerce. In Intelligent Information Agents, M. Klush (Ed. ), Springer Verlag 1999, p. 131 -152. P. Noriega, C. Sierra. Auctions and multi-agent systems. In Intelligent Information Agents, M. Klush (Ed. ), Springer Verlag 1999, p. 153 -175. E. Durfee, e. a. . Strategic reasoning and adaptation in an information economy. In Intelligent Information Agents, M. Klush (Ed. ), Springer Verlag 1999, p. 176 -203. W. Brenner, R. Zarnekov, H. Witting. Intelligent Software Agents, Springer Verlag, 1998, Ch. 6, p. 267 -299. 26

Agent systems references n n n Bargain. Finder - part of "Smart Store Virtual" by Anderson Consulting Jango - Netbot Inc. , Seattle, USA Persona. Logic - Reordan, Soresen, 1995 Software Agents Group, MIT Media Lab http: //agents. media. mit. edu/projects/ Kasbah - project of MIT Media Lab, Chaves, Maes, 1996 Tête-à-Tête - Guttman, Maes, 1998 Firefly - Shardanand, Maes, 1995 Firefly Networks (does not exist any more) Agent. Builder Auction Agents for the Electric Power Industry http: //www. agentbuilder. com/Documentation/EPRI/index. html Fishmarket - Noriega, Sierra, 1997 UMDL - University of Michigan, Durfee e. a. , 1997 Info. Sleuth http: //www. argreenhouse. com/Info. Sleuth/index. shtml Retsina http: //www-2. cs. cmu. edu/~softagents/retsina_agent_arch. html 27