Hi Paul -
I recently attended a presentation given by Reid Smith of I2KConnect (.com) at the SW Chapter AIIM meeting. He described how his AI-based platform is being used today to auto-classify and tag content for the knowledge management site sponsored by the Society of Petroleum Engineers. Very much in line with your comments.
As you mention in your excellent article, the key to success with recognition and classification is domain knowledge. Such knowledge can only be acquired and made re-usable as a result of iterative interactions between human domain experts and ML software that's working against large volumes of source information. Domain knowledge doesn't emerge spontaneously from simply doing pattern recognition of document contents.
I2KConnect can also provide domain expertise in the healthcare arena. Might be useful to you.
Good Luck,
- Mitch
------------------------------
Mitch Shults, RecordPoint Software
------------------------------