Email Filtering: A Case for Fast and Easy Learning

Early in Saffron’s life, a couple of guys, Bob Cagle and Dean Pfutzenreuter from Open Field software on the West Coast, understood the power of our associative memories and applied it to spam filtering.  Beyond just spam, their Electronic Learning Assistant (ELLA) was a personalized email management system for any set of folders defined by each user.  As reviewed in PC User Magazine when compared against many other solutions including collaborative filtering and a naïve Bayesian, ELLA was declared “World’s Best Spam Blocker”.  It was similarly praised by other reviews in Forbes and Fortune and by end users as “near perfect”.   As one user wrote, “[I] Have been using Ella for a few days. But, after only two mistakes so far, Ella has been 100%.  As they say, it learns as it goes. Seems to be the case.”

ELLA proves the incremental, non-parametric, nonlinear learning of associative memories.  Any user – even a cave man – can simply create a folder, show ELLA a few examples though a wizard interface, and it is off and running.  No “black art” parameter tweaking.  Learning on the fly.  Add new folders at any time or teach it new cases as spam attackers also change their tactics.  By simply correcting it when it makes a filing mistake, it keeps tracking to near perfect.  This is due to the highly nonlinear representation of the memories but also to the brilliance of Bob and Dean in how they represent email.  Bob and Dean are shining examples of our customers and partners as “creative entrepreneurs”.   Given our nonlinear engine to reason about attribute interactions, they developed the 100 email attributes that also served to make ELLA so accurate.  The power of the engine was combined with the brilliance of application design.

Over the years, ELLA provided awesome value to Saffron when I would talk to a room of potential customers and at least one person was a happy ELLA user, giving us immediate credibility.   ELLA is no longer sold but is available for free download.   I thought this was a long past story of our work in personalization – until this week:   A potential customer was looking around for new technology to solve a particular banking problem (something we also do) when he stumbled on ELLA.  He downloaded ELLA and loved it.  He then dug around to find us as the underlying engine, thanks to the nice comments on the Open Field Web site about Saffron as a partner.  We have the same high regard for them.

I thought to provide this history and these links to ELLA as an example of memory-based classification.   We are adding our REST “classifications” method to TweetDive as another example, which will be available soon.  Whether you are thinking of reducing email overload or any more serious problems, think of ELLA, blending a powerful engine with excellent application.  Even if you don’t use Microsoft Outlook, check out the wizard demo of how fast and easy ELLA can be trained.  Thanks to Bob and Dean for this one.  We had only a small engine, called SaffronOne, when we first met them years ago, which is why ELLA runs on a personal computer.  Now we have SaffronMemoryBase for massive scale on distributed systems.  On to the cloud!

Tags: classifications, ELLA, email

This entry was posted on Thursday, March 11th, 2010 at 11:17 pm and is filed under Natural Intelligence, Partners, SaffronMemoryBase. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.