Vivisimo Federal

1
(No Transcript)
2
About Vivisimo Inc.

• Enterprise software company founded June 2000
• Carnegie Mellon computer science dept spinoff
• Founding technology is document clustering
• Award-winning web-search site Vivisimo.com
• Showcases our clustering meta (federated)
  search products
• Best meta-search site 2001-2002 (Search Engine
  Watch)
• 2nd place this year, but powered 1 Dogpile.com
• Main funding National Science Foundation SBIR
• Also, project started under NSF Funding at CMU in
  1998

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

• Overlooking information has a business/user cost
• Employees fail to find needed information
• Citizens dont solve their problems by
  themselves, online
• Publishers lose potential readership
• Web search providers lose potential
  click-throughs
• Users miss out on unexpected discoveries or
  opportunities

4
How to Alleviate Information Overlook?

• Decrease the available information
• Purge the obsolete
• Censor the worthless
• Make people
• Smarter
• Work harder
• Read faster
• More efficient! (but how?)

• Best ROI Provide organized information!

5
How to Provide Organized Information?

• Manually tag (classify, index) all content?
• We have no process for consistently tagging our
  content. We have 50 different business units.
  People in one unit do a great job, but others do
  not use tags at all. Forrester Report
• Forrester says 4 per page to make a controlled
  vocabulary
• 50 per document to manually tag (large pharma)
• Tags tend to be broad and bland (one size fits
  all)
• ? Tagging is costly and leads to bland results

6
On-the-Fly Document Clustering

• Cluster top 200-500 Search Results at The Output
• Uses title, snippet, and (optionally) meta-tags
• Works with Autonomy, Convera, FAST, Google,
  Sharepoint, Ultraseek, Verity, etc.
• Folder descriptions are as fresh as your content
• No need to develop/maintain a taxonomy
• Long History of Research
• Industry IBM, Xerox, Verity, Microsoft
• MIT, Stanford, Carnegie Mellon, etc.
• What have they missed?
• Quality of description is the most important
• Cluster Descriptions Should Be
• Concise, accurate, natural, distinctive
• Cant optimize these in the traditional way
• Instead, invent a great heuristic algorithm
• Add lots of domain knowledge (English, German,
  etc. specific verticals)

7
Screenshots of Three Demo Govt Sites
8
Screenshots of Three Demo Govt Sites

• City of San Diego
• sannet.gov
• http//vivisimo.com/demos/SanDiego.html
• DOE National Energy Technology Laboratory
• netl.doe.gov
• http//vivisimo.com/demos/NETL.html
• Centers for Medicare and Medicaid Services
• cms.hhs.gov
• http//vivisimo.com/demos/CMS.html

9
(No Transcript)
10
(No Transcript)
11
(No Transcript)
12
Modeling the Value of Topic Clusters

• Intuition
• Lots of wasted effort if information is
  disorganized
• View few results before exhausting your patience
• Modeling Assumptions
• User spends 12 min before giving up or moving on
• Eye skips over search results or folders
  sequentially
• 1,000 employees at 60 per hour
• 2 searches per employee/day
• 10 minutes to solve problem elsewhere when search
  fails
• Folders let you see 11 docs in detail vs. 6 for
  ranked lists
• Conclusion savings of 1M per year (white
  paper)
• Fortune 1000 customer experimental tests
• Combination of meta-search and clustering
• Minutes-to-solve 1.9 versus 3.7

13
Customer Highlight

• Cisco Systems Technical Assistance Center
• Several thousand support engineers
• Search platform is Google Search Appliance
• Had evaluated taxonomy-building products
• Cisco has complex, rapidly evolving product line
• Effort involved didnt scale with the amount of
  information
• Rejected the taxonomy-building approach
• Problems Challenges
• Lack of organized information
• Ranking well is hard to do on an intranet
• Separate collections couldnt be cross-searched
• How to leverage rich meta-data?
• Author, date, product version, etc.
• Solutions
• Content Integrator meta-searches Google
  collections

14
Multiple Sources Multiple Searches and Multiple
Lists
15
A New Standard, A Better Way

16
How the World is Now
17
Example of What Vivisimo Makes Possible
18
Value Proposition of Clustering

• End-users
• Easy access to useful but low-ranked results
• Learn at a glance the types of available
  information
• See results in context of similar results
• Agency
• More efficient employees/citizen support
  operations
• Citizens become more engaged with your content
• IT Dept
• Quick installation on any search engine/database
• Overlays search is non-invasive no maintenance
• No need to train users

19
What We Ask Potential Customers

• What search engines do you use?
• Are you happy with it them? If so, proceed.
• If not, ask our opinion and/or inquire about our
  partnerships.
• Are you ready for clustering?
• Can your search engine deliver 200 results
  acceptably fast?
• Are the descriptions of acceptable quality? send
  a screenshot!
• Do a pilot (2-4 weeks)
• Customize with stopwords and thesaurus (optional)
• takes minutes or days

20
Customer Quotes
21
Vivisimo.com
Web search demo at Vivisimo.com showcases our
technology. ? 7M searches monthly ? 22M page
views

• International recognition
• Powered the 1 meta-search engine 3 years running
  Search Engine Watch Awards
• Better than Google? CNN.com
• Analyst Choice Award eWeek Labs
• Named Top 100 Company KMWorld and eContent
  Magazines
• Site of the Week PC Magazine
• Major Analyst Coverage Delphi, Forrester,
  Outsell, Seybold and IDC

22
Vivisimo Federal

• Vivisimo on GSA Schedule through Onix Networking
  Corp
• Onix Networking is a Google Search Appliance
  reseller
• Work with several DC systems integrators and IT
  consultants
• Several clients (govt intelligence) plus ongoing
  pilots

23
Conclusion

• Information Overlook imposes high opportunity
  costs
• Alleviate by showing organized info
• Clustering into folders is a natural approach
• Hard to do well but now a solved problem
• Overlays any search engine (Google, Autonomy,
  Verity, etc.)
• Organizes info at delivery time rather than
  creation time
• No taxonomy-building headaches

Info http//vivisimo.com/gov FirstGov clustering
demo http//vivisimo.com/firstgov Raul
Valdes-Perez (valdes_at_vivisimo.com)