About

1
(No Transcript)
2
About

• Vivisimo Inc. is an enterprise software company
• Creates innovative software to access and cluster
  the worlds information, for better search and
  discovery
• Founded June 00 in Pittsburgh, 20 employees,
  sustained profitability
• Organic growth, no venture capital funding
• About 80 customers Cisco, JNJ, NSA, JAMA,
  Micropatent, AOL, AAAS, etc.
• Clusty.com, acclaimed web search engine
• Launched on Sept 30 of last year
• Raul Valdes-Perez, CEO co-founder
• Last Stop Carnegie Mellon Computer Science Dept
  (1986-2000)

3
Problem Information Overload Overlook

• Information Overload and Information Overlook
• Look for information ? get too much back
• Most people handle overload by overlooking most
  information

• How to Get People to Overlook Less Information?
• Provide categorized information!

4
Categorization at Creation Time

• Also Known as Taxonomy Building
• Create a controlled vocabulary (taxonomy) of
  categories
• Index new documents into the taxonomy
• Update the taxonomy over time
• At search time, group results into most frequent
  categories
• Examples
• National Library of Medicines Medical Subject
  Headings (MeSH)
• Developed over decades
• Human indexers assign MeSH terms to new articles
• Ten people refine MeSH over time recently added
  SARS
• Yahoo directory, Open Directory
• Why Not Widespread in Electronic World?
• Taxonomy model is costly complicated
• End-users need to meta-search, which undermines
  taxonomies
• Much of electronic world still in the stack of
  books on the floor mode

5
(No Transcript)
6
(No Transcript)
7
(No Transcript)
8
(No Transcript)
9
New Way Clustering

• To cluster means to form groups
• Stars, candy, disease outbreaks, computers
• Quick linguistic/statistical analysis of search
  results
• Forms groups based on the major themes in the top
  N results
• Can leverage taxonomies, as shown at
  http//Clustermed.info
• Software based on 6 years of research
  development funded by National Science Foundation
• Builds on many research-years by others in
  academia industry

10
End User Benefits of Clustering
User Benefits 1 See what's available At a
glance, the folders show you an information
landscape. 2 See much further Following our
interests, we navigate to low-ranked but
interesting search results, which we're unlikely
ever to see otherwise. 3 See similar
information together We don't have to be
satisfied with the first reference we come
across. We can compare several and pick the best
one.
Key Advantages 1 Works on the fly No need for
pre-processing 2 Spontaneous categories no need
to pre-define them
11
Question Clustering, Query Refinement,
Personalization, or Entity Extraction?

• Query Refinement - Show alternative queries to
  the user
• Pro little computation, dont need search
  results as input
• Con
• History-based, not matched to search results,
  relevance is a challenge
• Click on a query, your screen disappears, absorb
  new context
• Personalization
• Pro its about you!
• Con
• Peoples interests arent static (Olympics,
  Oscars, Tsunamis, etc.)
• Shared computers lead to shared personalization
• Entity Extraction
• Pro nouns are informative and familiar
• Con
• verbs matter (and adjectives, adverbs, etc.)
• ungrammatical search-result descriptions
  
  

12
(No Transcript)
13
Question Index Everything or Meta-Search?

• Index Everything
• Pros centralization, universal ranking,
  simplicity
• Cons centralization, universal ranking,
  complexity
• Meta-Search
• Pros decentralization, can leverage
• Partners, secondary content, free/government
  search engines
• Ranking methods can leverage voting
• Publishers aggregators can create vertical
  destination sites (vortals)
• Cons
• Need to create relationships with partners
• Need meta-search software
• Dont have a single ranking score for everything
• Is Meta-Search a Practical Necessity?

14
(No Transcript)
15
Some Early Adopters of Clustering
16
Old Way of Thinking

• If a search returns many hits, then youre a
  novice
• Most patrons/users are novice searchers!
• New way
• Purposely start out with broad searches
• Learn something by viewing the results
• Use what you learned to search again
• Find what youre looking for
• or discover what youd normally miss

17
Clustering Will Go Mainstream in 2005-06

• Why? Critical Mass of Installations and Buzz
• AOL Clusty reach about 10 of web searches
• Unbeatable Value Proposition
• Instant organized information without taxonomies
• Can leverage taxonomies where they exist
• No surprises
• Real End-User Benefits
• Users can consider lots more info with the same
  effort
• Makes people smarter!
• Transform online world of information
• Less like 1-person used bookstore with piles of
  books
• Look more like Barnes Noble or Borders

18
(No Transcript)
19
(No Transcript)
20
(No Transcript)