Transcript Intro ACM SIGMOD

®
Microsoft Site Server
Commerce Edition
Jay Sauls
Microsoft Consulting
Services
Overview
Business Proposition
Solution Architecture
Shopper Experience
Technology Solutions
Business Proposition
Why go online?
Online
Requirements
Wider reach
High scalability
Reduced “friction”
Integration
Always open
24x7 availability
Solution Architecture
Internet
Windows Load Balancing
Service used for directing
requests to a web server
...
Web Server 1 Web Server 2
Web Server N
Microsoft Cluster Services
used for failover capability
on SQL Server
Database Server 1
Shared Disk array
Database Server 2
Design
Transact
Engage
Browser
Wallet
Authoring Tools Pipelines
• Order
Processing
Ad Server
• Commerce
Interchange
Personalization
Membership
Third Party
Components
Analyze
Order Analysis
Usage Import
Commerce
Reports
Windows NT Security, IIS, MTS
SQL, Oracle Database
Line of
Business
Application
Shopper Experience
Browse
Select
Purchase
Product in stock?
Valid address?
Special discount?
Valid credit card?
Pipeline Architecture
Browse
Select
Purchase
Product Info
Validate bill_to
Shopper Info
Validate CC
Sale Adjust
Inventory
Flag Inventory
Authorize CC
Make PO
SaveReceipt
SendSMTP
SQLOrder
Purchase Pipeline
Item Promo
Product Pipeline
Item Price
Pipeline Details
Validate CC
Order items
Make PO
Shopper info
SaveReceipt
Payment Info
SendSMTP
SQLOrder
Purchase Pipeline
Authorize CC
Transaction
Validate bill_to
Data Mining
Store owners need information
How many visitors?
How many buyers?
When do they shop?
What do they buy?
Where are they from?
Data Mining
Current Process
Web Server logs imported into SQL Server one row at a time
Each row is processed by a stored procedure
Data from rows checked against dimension tables
Only one import process can run at a time
Data Mining
Leveraging MS Research Technology
Realtime Recommendations
Expands Intelligent Cross Sell from 3.0
Value-added functionality specific components
Product cross-sell
Personalized product recommendation
Algorithms can be parameterized for confidence levels
Segmentation - Find interesting sub-populations for targeting
Label new user segments for advertising, promotions or direct mail
Update User profiles
Techniques
Explicit via enabling Business User using OLAP tools with Data
Warehouse
Implicit via Clustering Algorithms from MS Research
Clusters imply similarities in behavior or likely response
Data Warehouse Architecture
1
Event Data
4
Transactions
Interchange
IIS
3
Administration
ADO
Real Time
Events
Commerce
OLE DB Provider
5
DTS Packages
Data Warehouse Service
Win Media
Apache
Other Events
2
Non-Event Data
User
CSF
Import
Tasks
User
Usage
Transaction
CSF
Catalog
Interchange
Win Media
Content
Log Manager
Site Vocab
IP Resolution
CMD Proc
ROLAP Svc
Cube Mgr
Schema Mgr
6
SQL
Server
ADO
OLE DB
Catalog
WhoIs
Other Data
Custom Task
Data Mining
Design Points
Microsoft.com : > 100M hits / day = 30M – 40M “useful” hits
Use OLAP cubes to view imported dimension data
Number of dimension values is potentially unlimited
Data will drive new Recommendations (Predictor) models
Data Mining
Enhanced Design
Rows are imported in batches of 3K – 4K per batch
Batches are analyzed in-memory for distinct dimension values
Distinct dimension values are added to SQL Server in batch update
Multiple batches can run simultaneously
Data can be partitioned across multiple databases
Targeting Applications
Content
Separation of page logic, format and data
Allow a business user to easily manage, format and target content on a page without
the need for a developer
Allow developers
Simple programming interfaces for the VID developer
High performance and scalability (10ms/Slot)
Extensible formatting templates
Advertising
Campaign or Campaign Item impression goals
Exclusive targeting for sponsorships
Exposure limits
Support for Ad Networks (LinkExchange)
Discount Campaign Management
Reacts to product page and user’s basket
Related Sells Campaign Management
Supports Up-Sell, Cross-Sell and Inventory Sell
Direct Mail Campaign Management
Fast, Scalable, Runs as an NT Service
Has a List Management object to support importing and merging of lists
Campaign tracking of mails sent, clicked
Targeting Architecture
Design
Time
Content
Selection
Framework
Expression
Builder
GUI
Load Prediction
Model
Expr
Evaluator
Expr
Evaluator
User
Profile
Schema
Predictor
Engine
Store/
Retrieve
Expressions
Load Predictor
Data
Context
Profile
Site
Terms Datastore
Offline
Processing
Predictor
Client
Biz Desk
App
Context
Profile
Schema
ASP pages
Run
Time
Profile
Definitions
Expression
Datastore
Biz Design Data Store
User
Profile
User Profile
Datastore
Data
Warehouse
Recommendations
Architecture

scheduler launches model builder

app downloads current model

user goes to checkout page

ASP calls predictor to add recommendations

IIS logs feedback basket data to the warehouse
data
warehouse
OLAP
predictor
service
global.asa
model
basket.asp
IIS
model
Questions?