Archive

Posts Tagged ‘technology’

How does gwabbit gwabbit?

June 17th, 2009

How does gwabbit gwabbit? Automatic vs. manual contact recognition technology

gwabbit’s automatic contact grab capability represents a breakthrough in contact recognition technology, and a major departure from older generation manual contact grabbers. Unlike manual contact grabbers, which require the user to identify the contact by “painting” it with a cursor, gwabbit proactively scans email to automatically recognize contacts, then parses the component fields into their appropriate parts. This automatic recognition capability enables gwabbit to realize a product that quickly and easily grabs contacts without the time-consuming and tedious manual process of painting contacts. This is the key differentiator between a novelty product and a serious productivity tool.

gwabbit_popup

As you know, automatic recognition is not an easy thing to do. While there is presently no such thing as perfect recognition (100%), gwabbit employs a unique 2-method semantic approach to achieve its very high recognition rates.

 

How does it work?

Method 1: Character recognition.

This is a traditional approach, where the product identifies characters, letters, or words associated with certain fields, such as “Phone”, “( )” (associated with phone/fax numbers), and “@” (associated with email addresses).

 

Example 1: character recognition

Mr. Gwabbit Wabbit
Head Honcho
gwabbit
Phone: (555) 555 – 1111
Fax:
(555) 555 – 1112
gwabbit
@gwabbit.com

 


Method 2: Probability model.

This approach employs something called the“Hidden Markov Model” that uses probability to determine the placement of fields relative to other fields. For example, there is a high (but not perfect) probability that a job title field will follow a name field, but precede a company name field. One of the advantages of using the probability model is that it enables gwabbit to function with a remarkably high degree of success not only with American English, but with most other languages as well, including those employing double Byte character sets.

 

Example 2: Hidden Markov Model - probability model

Mr. Gwabbit Wabbit
Head Honcho
gwabbit
Phone: (555) 555 – 1111
Fax: (555) 555 – 1112
gwabbit@gwabbit.com


The combination of character recognition and probability models (fine tuned with thousands of case models) enables gwabbit to achieve a high degree of success. How high? On average, gwabbit automatically recognizes contacts well over 80% of the time. By using the “Improve Results” feature in situations where gwabbit doesn’t automatically recognize a particular contact (such as in threads with multiple contacts), this success rate improves to as high as 97%! Though 97% is a remarkable recognition rate, gwabbit will not be satisfied until we’ve achieved 100% automatic contact recognition.

 

To aid our efforts to refine gwabbit’s recognition engine, in gwabbit for Outlook we’ve added a feature called

“Feedback” to the gwabbit toolbar.


gwabbit_improve_toolbar_09

 

When clicked, Feedback sends the target contact to our developers for analysis. We routinely make free upgrades available to our customers which include the latest refinements to gwabbit’s recognition engine.

Features , , , , , ,