E2 Reader- Introduction

The E2 Reader is a Web app for testing ways to explore content that is highly interlinked. It uses the web of linked articles and people that is Everything2 as a testbed for graph-based discovery and recommendation techniques. It will:

  1. Try to replicate the basic E2 functions for search and discovery with a read-only app.
  2. Identify relationships in the existing link structure that are unused, under-used or used ineffectively.
  3. Identify new relationships that improve discovery and recommendation (curation).
  4. Experiment with user interest modeling and controlled serendipity for personalized discovery that avoids the 'filter bubble'.
  5. Test different interfaces for using the relationships to explore E2 and discover the gold hidden in its great bulk.

"That's all fine and well, but E2 does that already, you fool!"

Well, to some degree, yes, but I think E2 is currently using only about 10% of its brain, so to speak. I want to bring that up close to 100% and to later integrate the Web outside of E2 into the discovery process as well.

Premises for building discovery techniques:

  • Content has a large and complex context (lots of natural links to other stuff)
  • Content is discovered most naturally and efficiently when stored in a graph database structure that captures its full context of people and other content.
  • Graph structures facilitate discovery of implicit and emerging relationships.
  • Discovery by following relationships is fun and easy because that's the way the mind works.
  • Content discovery is more effective and efficient when driven by personal interests, aided by recommenation, and modulated by a degree of randomness. Serendipity, the joy of finding interesting things that are incidental or even unrelated to what is being sought stimulates exploration.
  • An effective UI for discovery uses the true (graph) data structure (linking).


Development will be directed in large part by user feedback.


*E2 Reader will try out personalized discovery and recommendation techniques and interfaces that emphasize search through exploration of a dynamically-structured, many-dimensioned semantic space guided by individual profiles of likes and interests. Controlled perturbation will be used to avoid the usual pit-traps of interest-directed search will also be evaluated.

Next in series: Implementation