The Future of Search Engines Take it personally! אינטרנט 2008 Emil Ismalon Co-founder and CTO Collarity, Inc. Internet 2008 February 25, 2008
My Tag cloud PhysicsBuddhistmeditation Quantum Holistic approachmeasurement problems Machine learningcognition & reality Zen The forceAbstractions Loveinformation retrieval mathMind& matter Lia NoaCOLLABORATION Multidimensional clustering basketball Social networkinghow? DRUMS Collarity high frequency lasers Swimming dear mama
Beyond the Limits of Keyword Search Web 4.0 The Semantic Web Syndicated WebIntelligent Web Web 3.0 2010 - 2020 Automated Content Analysis Productivity of Search The Social Web User Modeling Web 3.0 Web 2.0 2000 - 2010 The World Wide Web 2010 - 2020 Web 1.0 Naturallanguage search Universal/Open IDUser profiling 1990 - 2000 Tagging The Desktop PC Era Keyword search 1980 - 1990 Directories Files & Folders Databases Amount of data ** From: Making Sense of the Semantic Web, BY Nova Spivack
The Next Steps Today’s Content Social networking Media consumption Commercial consumption TechnologicalEnablement Methods Semantic Web User Modeling Automated Content Analysis Latent Space construction (Augmented LSA LDA methods) Universal/Open ID User Profiling Behavioral & Intentional Targeting User interactions with content
Projection into the future • Better rankings-relevant results • Recommendation engine • Deep personalization • Digidentity
Identity Digital Entity User Avatar Digidentity “What I’m about” My digital imprint Independent Entity Shabti statues
Key Element: PERSONALIZATION
The pros & cons of personalization Pros: • This is how the “real world” behaves: intents and meanings can’ t usually described by a query • It is the natural way to achieve higher relevance and enhanced user experience • Think about web navigation in 5-10 years from now.. Do you really think intelligent personal web agents will not emerge?
The pros & cons of personalization Cons: • Lockup problems • Cold start - it takes time and education • Is there a general solution to personalization or is it strongly dependent on contexts • Privacy ethics • Manipulations
Gradual perception of personalization • Language preferences • Geo targeting • Simple Lingual Disambiguation • The case of Java; is it programming OR island OR coffee? • Social networking • Form of preferable media • Textual OR visual? • Summary OR deep analysis? = =
Gradual perception of personalization • Understanding the right perception while interacting: • Pro OR anti • Believer OR skeptical • Subtle ones • Quantum physics article most suitable to a biologist
User modeling Eleanor Rigby’s recent queries: Paul is dead Did we land on the moon? Elvis is alive! Who murdered JFK?
ABSTRUCTION Collarity live example of conspiracy: Moon Elvis JFK http://www.collarity.com/search?query=Moon+Elvis+JFK
Natural Audience Segmentation After: Natural Communities Before: Traffic • What are my important segments doing? • What are their likes/dislikes? • Are they finding what they’re looking for? • How are their tastes changing?
The ability to personalize In order to materialize the subtle level of personalization an holistic approach is needed! understand the user from many different aspects: • Implicit Informational networking • Segmentation Cluster users in information space • Topic based communities Clustering users around topics
Space & Beyond • Establishing the “Space” • Live metric - relevancy coordinates (attributes) • Embedding users profile • Space dynamics • Predictions – Segmentation, Clustering
Future of DigidentityDigidentities Spontaneously Behave User News Work Social TV Hobbies Music Interests Knowledge
Conclusions and Summary • Personalization involves “intelligence” characteristic; abstraction, generalization, predictive clustering • Users implicit informational communities can serve as a natural enhancement of user searching and browsing modeling • Enable Users to control personalization features and variables
Further readings • Pros & cones personalized searchhttp://searchengineland.com/070309-081324.php • Dynamic Social Network Analysis using Latent Space Modelshttp://books.nips.cc/papers/files/nips18/NIPS2005_0724.pdf • Constraint-based Personalization Model: Multi-Channel Messaginghttp://www.research.att.com/~rjana/TothNagboth.pdf • A Hybrid Web Personalization Model. Based on Site Connectivity. • Miki Nakagawa, Bamshad Mobasher ... on non-sequential models, such as association rules and ...http://maya.cs.depaul.edu/~mobasher/papers/NM03b.pdf • Social Balance Theory • Revisiting Heider’s Balance Theory for many agentshttp://www.au.af.mil/au/awc/awcgate/lanl/social_balance_0405041.pdf