Case-Based Reasoning in E-CommerceWhat is E-Commerce?How can CBR help?Slide 4What’s wrong?We have a problemSlide 7Some Preliminary InfoIndividual Wish PropertiesSlide 10Overall Wish PropertiesProduct ClassificationsHow Do These Properties Help?Transaction ModelPre-SalesSlide 16ExampleHow Does It Work?SalesNegotiationSlide 21Slide 22Slide 23After-SalesSlide 25SummaryReferencesCase-Based Reasoning in E-CommerceJoe SoutoCSE 435What is E-Commerce?“The exchange of information, goods, or services through electronic networks”1How can CBR help?How many times have you seen this?How can CBR help?Or this?What’s wrong?Demand is either over-specified or under-specifiedIt is up to the user to find what they wantThere is no intelligent sales supportWe have a problemBuyer has limited knowledge of product baseSeller has limited knowledge of buyer’s requirements”Knowledge Gap”We have a problemKnowledge gap is solved in real-life by a human sales agent as a mediator. We don’t have this luxury online.Solution: CBR approach  product knowledge is stored as experience in a case base. Sales agent makes recommendations based on the stored experience.Some Preliminary InfoWe need a way to define user requirementsCustomers buy items in order to satisfy their desiresDefine a customer’s desire as a “Wish”Wishes have various propertiesIndividual Wish PropertiesImportanceHard: MUST be met (ie: “vacation for <$2000”)Soft: not essential, but helpful (ie: “red” car)Agent must satisfy ALL hard req’s and as many soft as possiblePrecisionPrecisely Determined (specific, ie: “>3GHz P4”)Undetermined (vague, ie: “fast processor”)Individual Wish PropertiesCertaintyCertainUncertainSales agent must try to increase certainty of wishes and make recommendations based on themOverall Wish PropertiesRedundancyWishes can be redundantEx: Computer that’s “fast” and can play Half-Life 2Agent must recognize and avoid redundant inquiriesConsistencyWishes can be contradictory Ex: new Ferrari, and under $1000Agent must either ask user to clarify, or suggest products that satisfy one of the two wishesProduct ClassificationsHow Do These Properties Help?1. Customers want a product to satisfy a wish2. Products have various properties3. Therefore, product properties can be mapped to the satisfaction of a customer’s wishWith all that in mind, now we can look at the transaction processTransaction ModelSingle transaction can be modeled with three phasesPre-SalesBuyer wants a product, Seller provides information3 PhasesSupplier SearchClient determines which supplier can satisfy their wishesProduct SearchMapping of customer criteria to productsNegotiation1. Price and way of payment 2. Details of delivery 3. Regulations about cost and deliveryPre-SalesRecall the Google ExampleNo “intelligent sales support”Burden of knowledge is in hands of the customersExampleDue to Knowledge Gap, Analog Devices added a CBR system to assist Pre-SalesAnalog Devices:http://www.analog.comHow Does It Work?Similarity Metrics!Similarity function for single attributeOK to be under, less similar if over desired valueThe overall similarity is computed weighted average of local similarities.Remember the “priority” boxSalesProduct has been chosen, must be configured and paid forCustomer and Sales Agent negotiate about product attributes and costs Intelligent Support is needed for negotiationNegotiation“A process where two parties bargain resources for an intended gain”1In Sales phase, customers navigate through products to satisfy their wish. Some wishes known, others discovered in the process. Hard wishes must be fulfilled, soft wishes can be negotiated. Agent finds out these demands with the customer and finds a product which fulfills them. Agent can be “Active” or “Passive”SalesCBR Model must be modifiedStandard Model:2. Reuse3. Revise4. RetainCase Library1. RetrieveBackground KnowledgeSalesNew ModelNo Retain phase: sale does not add another product to the product baseAdd Refine phase: user demands refined based on the evaluations given by the customer.ExampleCBR approach to negotiating a BMW saleAgent here is passiveButtons for “sportier”, “more comfortable”, “cheaper”, etc.After-SalesCustomer has already bought a product and needs support during its usageTo assist the customer, they are supported with a case base of possible product problems, a query interface, and similarity measures which should help to find a similar problem and solutionMany companies have online CBR customer-support websites (Dell, 3Com, etc)  Help Desk SystemsExampleDell Support site:http://support.dell.comSummaryE-commerce is a growing field with lots of potential revenueStandard search technology is too limitedCBR can be applied in all 3 transaction phasesKey is to provide intelligent sales support  agent guides customer through each phase of transactionReferences1. “Intelligent Sales Support with CBR”Wilke, Lenz, Wess2. “Experience Management for Electronic Commerce”Bergmann3. Wikipedia: