Discussion Report 27 Mark Hurst - Semantic Profiling and Social Networking.

Title of Session

Semantic Profiling and Social Networking

Convenor

Mark Hurst

Participants

Craig Duncan, Marc Storms, Chris Addison, Dale Chadwick

Key Discussion Points

Introduction

The aim of this session was to introduce semantic profiling technology and brainstorm the areas where it can be useful to the KM4Dev community. Semantic profiling uses automatic tagging and context building techniques to extract implicit knowledge from text in websites, emails, blogs or documents etc. Discovery tools can help with expert location and improvised team building. Exploration tools can help cluster related organisations or individuals around shared context. Knowledge mapping can help with project profiling and comparison.

Are people thinking the same?

• It would be useful to cluster organisations around shared context.

Idea tracking

• May be possible to track ideas through the network.
• Requires support for time dimension in visualization tools.

Bias and weighting

• Must support bias control by using profile weightings since the South may have a relatively small amount of data.

What are the corporates using for profiling and exploration?

• Collexis - "expert profiling, combining knowledge management and discovery with social networking"
• FastSearch - "automatic entity extraction"
• Endeca - "search, information access and guided navigation solutions"
• Autonomy - "meaning based computing"
• Flamenco (Berkeley) - "explicit exposure of category metadata, to guide the user toward possible choices, and to organize the results of keyword searches"
• Google Appliance

Who might use improvised team building for disaster or crisis management?

• Look at ReliefWeb
• International Strategy for Disaster Reduction - at early stages of knowledge and community building

Ontology questions

• How do you handle taxonomy differences between domains?
• Discussion of when ontologies should be applied, before or after extraction of natural entities.
• An argument was made for data to be kept in natural form and personal viewpoints supplied by applying ontologies on-the-fly wherever possible, since this supports pattern matching and keeps data from going stale. However, stopwords and forced themes should be available. It may also be helpful to provide more focus on entities rather than concepts.

Discussion of "context heat"

• The emergence of new contexts can be identified by the appearance of relationships between concepts without overlapping neighbours.
• This concept of 'emerging context' is the same as a 'weak tie' in a social network.
• Comparison with how the brain works when introducing new concepts to a language. Eg "mobile" and "phone" originally had no overlapping neighbours but are now in transition through "mobile phone" towards simply "mobile", with appropriate shared neighbouring concepts in the semantic network of English. See also Words for Change competition!
• Time dimension requirement identified for 'context emergence'. Also see Numenta and Jeff Hawkins' model of neocortex!
• Comparison with Flickr's "interestingness" index calulated using the followers of supernodes

Also see

• Drupal, BuzzMonitor, DevelopmentC