We envision this project to include, but not be limited to:

TalentMap is an employee survey company that provides insight to organizations about the happiness and specific concerns of their overall workforce.  As part of our surveys we ask employees to provide specific text feedback about issues of concern.  Currently this feedback requires human interpretation to extract useful information.

Our goal is to automate this process to the extent possible.  To do this we want to be able to do two things:

a) Identify and group comments that represent shared topics of concern in the answers to specific questions, and to find out what percentage of comments related to that topic.  The goal is to identify and rank the issues of greatest concern to their workforce, ideally without human intervention.  We would want a list of comment groups that were somehow similar, and a human would have to look at several of those comments to find what the actual subject was.  This isn't a simple classification job: some of these topics are specific to particular organizations (e.g. you might get an org that has just gone through some kind of restructuring and many of the comments are related to that, but you wouldn't necessarily be able to train a classifier to identify a "restructuring" topic as the language used varies significantly from org to org and industry to industry).

b) Identify high value comments within the list of comments that respond to a particular question.  As an example, some comments are short ('good job') some are very long when an unhappy employee writes a screed and some are simply not actionable ('everything sucks').  Comments of value to the organization contain specific information ('the third floor offices are too hot').  We would like to be able to separate low and high value comments so that human review was done only on the high value comments.