Transcription
Predictive Analytics overOn-line and Social Network DataUsama Fayyad, Ph.D.CEO, Open Open InsightsCopyright 2009 by U.M. Fayyad, All rights reserved0
AcknowledgementSome of the work presented in This talk was done informer position:Chief Data Officer & Executive VPYahoo! Inc.Some of this material extracted from prior talk:The Evolution of Search Technologyand the Role of Social Networking in MarketingUsama Fayyad, Ph.D.CEO, Open InsightsOpenResearchInsights1
Background and Lessons Grad student forever at University of Michigan, Ann Arbor– after 5 degrees, Lesson: real world can be more exciting Jet Propulsion Lab, California Institute of Technology (1989-95)– Data mining can help the highly specialized tremendously Microsoft (1995-2000)– The real problem is not data mining technology, but operations digiMine Inc. (2000-2003)– [now Audience Science]– The problem is explaining the strategic value of Data to exec team andconnect with business strategy DMX Group (2003-2004 )– Data strategy in business intelligence and delivering on the deepervalue Yahoo! (2004 – 2008)– The Internet is a much deeper social and commercial transformationthan I ever imagined Open Insights (October 2008 - )– Back to Data Strategy and Data Mining2Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Overview Yahoo! and On-Line Marketing Primer From Search/Advertising to Marketing:– The Evolution of Search and search marketing– More focus on Search Marketing Case-study: Impulse (Search-after) Case Study: Behavioral Targeting Case Study: Retargeting Case Study: Choicestream ad personalization The Evolution of “Social Media” (Introduction) Wisdom of the Crowds Case Study: Social Targeting at Yahoo! Case Study: Social Targeting using Twitter (VH1 example) What Does this mean for Search?– Where does Search evolve to? Concluding Thoughts3Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Globally, Internet Users Outnumber 1 Billion by 2007Internet Users in Millions:Worldwide TotalJapanRest of WorldWestern EuropeAsia/PacificUnited .0702.4602.4506.3400.0200.00.02001Source: IDC, December 2003.42002200320042005Copyright 2009 by U.M. Fayyad, All rights reserved20062007www.open-insights.com
Yahoo! is the #1 Destination on theWebMore people visitedYahoo! in the pastmonth than:73% of the U.S. Internet population uses Yahoo!– Over 500 million users per month globally! Global network of content, commerce, media, searchand access products 100 properties including mail, TV, news, shopping,finance, autos, travel, games, movies, health, etc. 25 terabytes of data collected each day andgrowing Use couponsVoteRecycleExercise regularlyHave childrenliving at homeWear sunscreenregularlyRepresenting thousands of cataloged consumer behaviorsData is used to develop content, consumer, category and campaign insights for ourkey content partners and large advertisers5Copyright 2009 by U.M. Fayyad, All rights reservedSources: Mediamark Research, Spring 2004 and comScore Media Metrix, February 2005.www.open-insights.com
Yahoo! Data – A league of its own Terrabytes of Warehoused DataMillions of Events Processed Per Day14,0005,0001,000500YSMY! GlobalY! MainwarehouseNYSEWalmartVISA100Y! PanamaWarehouseSABRE94Y! ND CHALLENGE PROBLEMS OF DATA PROCESSINGTRAVEL, CREDIT CARD PROCESSING, STOCK EXCHANGE, RETAIL,INTERNETY! Data Challenge Exceeds others by 2 orders of magnitude6Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
What we needed at Yahoo! Data Strategy Prioritization conducted based on business need,not IT– Drove new innovations in DB– Data segmentation management– Specialized stores– File systems for grid computing (Hadoop) Had to resort to own home-grown data storesfor largest applications7Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Advertising: Brand and DRKnowledge of users & their behavior throughout the purchase funnel cangrow brand & direct response revenueAwarenessConsiderationPurchase 220B BrandAdvertising MarketMost time & activity is in consideration &engagement, but there are limitedmetrics & reach strategies 250B DirectMarketing Market8Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
www.open-insights.comA Tale of Two Search EnginesOpenResearchInsights9
Algorithmic results Audience- 10Advertisements Monetization Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Algorithmic vs. Ad Search Analogous to classical separation of editorial vscommercial content Technical underpinnings:– Some commonalities (IR, ML)– Many differences (incentives, spam, mechanism design) Will cover both algo and ads11Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Search advertisingwww.open-insights.comOpen Insights12
Search queryAd13Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
A Question for the audience Do you think an “average” user, knows the difference betweensponsored search links and algorithmic search results Do you think an “average” user knows there are sponsored linkson the page? Do you think a user knows where a sponsored link wouldnavigate to upon a click?14Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Ads goin slotslikethese15Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Higherslotsgetmoreclicks16Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Economic ordering Yahoo!/Overture started with Bid Ordering of search adsThe only business model at the time Google introduced revenue ordering to avoid the Overturepatent– Resulted in 4x – 6x revenue per search! Bid and revenue ordering: two forms of ordering by an econscore– Auction mechanism matters (Vickrey Auction)– Is not the right generalization to Search Ads (Edelman, Ostrovsky,Schwarz 2006) Does revenue ordering maximize revenue? No – advertisers react to ordering scheme, by changingtheir bid behavior! Lahaie Pennock ACM EC 2007– Family of schemes bridging Bid and Revenue ordering– Game-theoretic analysis17Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
A new convergence Monetization and economic value an intrinsicpart of system design– Not an afterthought– Mistakes are costly! Computing meets humanities like never before –sociology, economics, anthropology We call it Computational Advertising18Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
A question for the Audience:Why is search-relatedadvertising so powerful?Open Insights19
Moving Customers up the FunnelImpulse Banners– Target users based on their activity – both search andproperty -- within the NEXT HOUR Behavioral Categories – Apparel, Computers, HomeAppliances – all the same categories that you can use forregular behavioral t 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Impulse Banner Example1Usersearches onthe word“CreditCard”All within1 HOUR2Sees that “CreditCard” falls under3the category:“Finance/Creditand CreditservesServices” “Finance/Credit andCredit Services”banners to Useranywhere on theYahoo! networkwithin 1 HOUR2125%- 261% higher CTR than RONCopyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Way Impulse Works Searches are not at all associated or trackedthrough personally identifiable information No long-term memory of search terms, allstored on client cookie. We generalize the category is targeting is atgeneric category: e.g. Financial Services, not“credit card” All targeting done in anonymous mode22Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
BT2.0 vs. Impulse (test data) Shoppers Testing (18 tests across eBT2.0 can utilizeHealth & PharmaSearch Behavior Data toLife Stagesdeliver higher effectiveness, RetailSMB & B2Band much larger inventorySportsthan targeting by SearchTechqueries only!TravelAverageCTR LiftReach lift 475.8%OverOverCTR LiftImpulseImpulse 291.5%171%254%Conv 16%984%222%429%129%867%59%246%144%652%*Note: these are straight averages rather than weighted averages. Therefore, might not be the best representation of the true overall CTR lift.23Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Behavioural Targeting (BT)SearchContentBTSearch ClicksAd Clicks24Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
What‟s „Behavioural Targeting‟? Targeting your ads toconsumers whoserecent behaviorsonline indicate thatyour product categoryis relevant to them.25Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Moving Down the Funnel New generation marketing solutions totake brand advertisers down themarketing funnelAwarenessConsiderationPurchase26Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
What goes into user behavior models?Relevant behaviors Pageviews Ads clicked Search queries 27AUTOS / SEDANBehavioural Interest SedanSearch clicksCopyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
How it works Network Interests yclesAnalyze predictive patterns for purchase cyclesin over 100 product categoriesIn each category, build models to describebehaviour most likely to lead to an ad response(i.e. click).Score each user for fit with everycategory daily.Target ads to users who get highest ‘relevance’scores in the targeting categories28Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Recency Matters and So Too IntensityActive now 29 and with feelingCopyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Differentiation Category specific modellingIntense Click Zoneintensity scoreExample 2: Category Travel/Last MinuteIntense Click Zoneintensity scoreExample 1: Category AutomotivetimeAlt Behaviour 1: 5 pages, 2 search keywords, 1 search click, 1 ad clicktimeAlt Behaviour 1: 5 pages, 2 search keywords, 1 search click, 1 ad clickDifferent models allow us to weight and determine intensity and recency30Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Differentiation Category specific modellingExample 1: Category Automotiveintensity scoreuser is in the Intense Click ZoneIntense Click Zonewith no further activity, decay takes effecttimeAlt Behaviour 1: 5 pages, 2 search keywords, 1 search click, 1 ad clickDifferent models allow us to weight and determine intensity and recency31Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Audiences Defined by Interests and ActionsExample: MortgagesResults from a client campaign on Yahoo!Network 626%Conv LiftWe found:1,900,000 people looking formortgage loans. 122%CTR LiftExample search terms qualified for this target:MortgagesHome Loans RefinancingDitechExample Yahoo! Pages visited:Financing section in Real EstateMortgage Loans area in FinanceReal Estate section in Yellow Pages32Date: March 2006Source: Campaign Click thru Rate lift is determined by Yahoo! Internalresearch. Conversion is the number of qualified leads from clicks over numberof impressions served. Audience size represents the audience within this behavioral interestcategory that has the highest propensity to engage with a brand or product and to click on an offer. www.open-insights.comCopyright 2009 by U.M. Fayyad, All rights reserved
Products: RetargetingRetargetingAn evente.g. People who viewed a certainad Target customers based onany past event Lead consumers down thepurchase funnelNext visit on Yahoo! Can be quite clevercreativelySecond ad with additionalinformation increases awareness,engagement!33– Improving offers There are time limitationsCopyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Example: Retargeting – BMW in GermanyBMW ran a homepage event in GermanyCan re-target those who saw the event, clicked on the event orsearched for BMW or a combinationCan increase conversionCopyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Brand Ads and Search Ads Interact! Is ad search strategy enough for a directmarketer? Do brand ads play a role in search advertising? Harris Direct Case StudyAwarenessConsiderationPurchase35Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Case Study: Harris DirectViewing These Ads:Had This Effect On: Aided Brand Awareness– Up 7% Brand Favorability– Up 32% Purchase Intent– Up 15%On:36Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Case Study: Harris DirectPeople who saw display ads were 61% more likely to search on related topics and drove 139% more clicks onalgorithmic and sponsored links specifically driving 249% moresponsored search clicks and driving 91% more activityon the HarrisDirect.com website.37Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Personalization Case-StudyOverstock.com Case Study
Personalization Drives Performance EverywherePersonalization Effects on Consumer MarketingAcross All Retail ChannelsAcquisitiRelevantonAdvertisingDisplay AdvertisingSalesRetentionPersonalized Up Sells& Cross-sellsPersonalizedRe-TargetingProduct DetailDisplay Ads Landing PageBrandPersonalizedLoyaltyBrand ExperienceBrand AdvertisingPersonalized Home PageBanner AffiliatesShopping CartEmail & Subject Line OffersLanding PagesOrder ConfirmationCopyright 2009 by U.M. Fayyad, All rights reservedCall Center & In-Store KioskProduct Recommendationswww.open-insights.com
Consumer Value Drives Business PerformanceRecommendationsProduct PromotionsDiscovery, Serendipity„More Like This ‟ ComplementsResults:400% ROIin 5 mos.Results:50% longerrental queuesCommerceCommunityOptimized „Next Sell‟ & „Cross Sell‟Like-Minded Buddies& Social GroupsResults:Results:300% longersubscriberlifetime250% more salesper visitDirect MarketingTargeted AdvertisingTargeted Mail & PromotionsPersonalized Ad DeliveryResults:30% Increase inVOD sales40Copyright 2009 by U.M. Fayyad, All rights reservedResults:550% higherrevenue perimpressionwww.open-insights.com
Overstock Showed Every User the Same AdMarciaFran Bought comforter last week Registered for baby items Abandoned cart with coat Recently bought a coat Lately browsing jewelry Added shoes to wish listCopyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
ChoiceStream Personalizes Each Ad MarciaFran Bought comforter last week Registered for baby items Abandoned cart with coat Recently bought a coat Lately browsing jewelry Added shoes to wish listCopyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Giving Shoppers a Compelling “Call to Purchase”Personalized Landing PageContains more product recommendationsFranClicks on product in adCopyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Improving the Metrics that Matter MostOverstock Cares AboutChoiceStream RealRelevance Advertising Lift*Conversionsper Thousand ImpressionsOver 3xRevenueper Thousand Impressions3x-5x* Source: Actual results based on A/B tests comparing advertiser's best-performing display ads against ChoiceStreamads with personalized product recommendations.www.open-insights.comCopyright 2009 by U.M. Fayyad, All rights reserved
www.open-insights.comSocial Mediaand the wisdom of the crowds.OpenResearchInsights45
Evolution of Social Media Although the “traditional notion” of portal andweb content is still attracting growing audiences The original notion of “publishing content” toattract audiences is changing fast– As people discover the fact that the Internet is anInteractive Medium– The uses of the Internet enter areas we could notimagine a short time ago A new notion of “publishing” is fast emerging– The opportunity of user-generated content46Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Examples of Social Media Blogs– The individual as publisher– Comments and tags part of the process Sharing Photos: e.g. Flickr Social Search– My Web 2.0– Yahoo! Answers– Del.icio.us Web communities:– Yahoo! Groups– Individual web presence: Facebook, MySpace, Yahoo! 360,Friendster, Video sharing: You Tube, Yahoo! Video, etc 47Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
48Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
What makes Flickr special?1. User Generated ContentContent not licensed from providers such as Corbis or Getty, but rather contributed byusers.49Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
What makes Flickr special?2. User Organized ContentContent is tagged, described, organized, discovered, etc. not by “editors” but by theusers themselves.50Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
51Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
What makes Flickr special?3. User Distributed ContentFlickr achieved distribution across the internet, not through “business deals” per se, butrather through the Flickr community which distributed Flickr content on 3 rd-party blogs.52Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
What makes Flickr special?4. User Developed FunctionalityFlickr exposed APIs (PHP, Perl, etc.) that allowed the community of developers to buildagainst the Flickr platform.53Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
What makes Flickr special?1. User Generated ContentContent not licensed from providers such as Corbis or Getty, but rather contributed byusers.2. User Organized ContentContent is tagged, described, organized, discovered, etc. not by “editors” but by theusers themselves.3. User Distributed ContentFlickr achieved distribution across the internet, not through “business deals” per se, butrather through the Flickr community which distributed Flickr content on 3rd-party blogs.4. User Developed FunctionalityFlickr exposed APIs (PHP, Perl, etc.) that allowed the community of developers to buildagainst the Flickr platform.Entire ecosystem created by less than ten employees aided by millions in the Flickr community.54Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
New Science? The Internet touches all of our lives:personal, commercial, corporate,educational, government, etc Yet many of the basic notions we talkabout:– Search, Community, Personalization,Engagement, Interactive Content, InformationNavigation, Computational Advertising– Are not at all understood, or well-defined– These are not disciplines that academia or anyindustry research labs focus on 55Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Areas of Research Community:––––How do you know what to believe on the Internet?Trust models on-line and trust propagationWhat makes communities thrive? Whither?Social media, tagging, image and video sharing Microeconomics: a new generation ofeconomics driven by massive interactions– Auction marketplaces– The web as a new LEI of activities and economies Information Navigation and Search– We are in the early days of search and retrieval Computational Advertising56Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
“Wisdom of Crowds”Is the Turing test always the right question?Open Insights57
A Digression:Computer Vision is hard58Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
59Copyright 2009Universityby U.M. Fayyad,All rightsreservedLuis Von Ahn, Carnegie-Mellon(GoogleImage)www.open-insights.com
60Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
61Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
62Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
63Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
What makes Flickr special?2. User Organized ContentContent is tagged, described, organized, discovered, etc. not by “editors” but by theusers themselves.64Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
65Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
66Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
67Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
68Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
69Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
70Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Challenges in social media How do we use these tags for better search? What‟s the ratings and reputation system? How do you cope with spam? The bigger challenge: where else can you exploitthe power of the people? What are the incentive mechanisms?71Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
www.open-insights.comCase Study: Social TargetingIllustrating New Directions on Yahoo! NetworkOpenResearchInsights72
What is Social Targeting?Social Targeting (ST) extends individual targeting techniques by consideringindividual characteristics of a user’s close social contacts.Individual Targeting Social Networks Social l73 Copyright 2009 by U.M. Fayyad, All rights ll of the above datafor friendswww.open-insights.com
Why use Social Targeting?Social Targeting provides indications of unobservableand off-network interest categories.Illustrative Social GraphPeople demonstrating interest inFootball by playing Yahoo FantasyFootballFriends with shared interest inFootball, but who don’t play FantasyFootball at Yahoo74Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Why does Social Targeting work?People share similar interests andpreferences with their close social contacts.HomophilyPeople seek out others with similar interestsDiffusionPeople within a group are exposed to similar ideasSocial IdentityCultural preferences signal group memberships75Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Yahoo! data for Social TargetingYahoo!’s social networks can help improve monetization and user experiencethrough better ad/content targeting.Fantasy SportsInstant Messenger76Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Empirical Tests Demonstrate Value of STTargeting Ads at custom lists of users based on Social Extension increasesinventory without decreasing CTR77Instant MessengerInventory Expansionfrom Seed PopulationCTR of Social Extension vsSeed Population 483% from sample ofBT Weight Loss117% vs sample ofBT Weight Loss 68% from sample ofBT Food & Nutrition104% vs sample ofBT Food & Nutrition 36% from sample ofY! Fantasy Football users107% vs sample ofY! Fantasy Football usersCopyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Case Study: TWITTER Social Marketing?Viacom’s VH1 Twitter campaign on ANVIL (the movie)(week of May 14th , 2009 – see AdAge article)Marketing ANVIL movieVery niche audience, how do you reach them?DiffusionGive 20 free DVD’s to major related artists/groups, ask themTo notify Twitter groups – reached over 2M peopleSocial Identity: power of word of mouth.What is the Cost to VH-1?Compare with traditional approach: TV commercials to promote adocumentary film?78Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
www.open-insights.comWhat Does this all Mean for Search?Many, many challenges.OpenResearchInsights79
Content trends[Ramakrishnan and Tomkins 2007]80Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Metadata trends[Ramakrishnan and Tomkins 2007]81Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
www.open-insights.comTowards Getting Things Done vs. SearchingOpenResearchInsights82
ExampleStart83I want to book a vacation in Tuscany.Copyright 2009 by U.M. Fayyad, All rights reservedFinishwww.open-insights.com
Loved the vacation, want to make that sweetItalian coffee at home85Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Trends in task complexity Dawn of search:– Navigational queries– Pockets of information Today:– Increasing migration of content online– New forms of media only available online– Infrastructure for payments and reputationsufficient for many users86Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Things to notice Long-running user goals Search as hub:– start there– return for resource discovery and at taskboundaries– traverse the web broadly to complete task Web services integrated into task87Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
What does this mean to Search? Publishers and search engine collaborate Users see richer search experience Accomplish their tasks faster and moreeffectively Example: abstracts surfacing structuredcontent Back to the Future?88Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
An Example: Search results in Search Monkey89yelp.comGawkerbabycenterNew York TimesepicuriousLinkedInanswers.comwebmdCopyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
An Example: Search results in Search Monkey90yelp.comGawkerbabycenterNew York TimesepicuriousLinkedInanswers.comwebmdCopyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
What are the challenges? Community of users– Social system Incentives and reputations– Economic system Poorly phrased, grammatically limited queries– Language analysis Improving user experience from past data– Data mining91Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
What we did not cover today. Applications of On-line Data for Enhanced Prediction– Product Sales (e.g. Car Manufacturer)– Customer Churn in Telco– Banking and Financial Services– Retail Cross-sell and Up-Sell The new generation of On-line Marketing Solutionsbased on the evolution of Search– Task orientation– Long-running queries92Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Experience summary at Yahoo! Dealing with the largest data source in the world (25 Terabyteper day) Search is in its early incarnations still. where will it evolve to?We don’t know– Verticalization trend– Solving tasks not retrieving pages– Back to the future: editorial input and directory V2.0?– Objects and entity extraction We cannot separate Search from Search Advertising– Think of the two search engines as intertwined andinseparable93Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
Lessons Summarized Evolution from Search and Analytics to Consumer relevanceand advertising On-line marketing is ripe with opportunities for predictiveanalytics applications Results are very powerful and we lack a true understanding New generation using social media data and more advancedpredictive analytics are very promising Combining context analysis (including text analytics) with usermodeling/targeting leads to some very powerful advances inrelevance94Copyright 2009 by U.M. Fayyad, All rights reservedwww.open-insights.com
www.open-insights.comThank You!Open Insights&Questions?Usama@open-insights.com95
Predictive Analytics over On-line and Social Network Data Usama Fayyad, Ph.D. CEO, Open Insights . .com. Open Insights Some of this material extracted from prior talk: The Evolution of Search Technology and the Role of Social Networking in Marketing Research Usama Fayyad, Ph.D. CEO, Open Insights 1 . The Evolution of "Social Media .
