A Data Science Central Community
In the applications section I would add also:
-document auto classification
-document auto extraction (automatic extraction of business attribute)
- pattern recognition (faces, stamps, ...)
-knowledge aggregation (via clustering methods)
In the subject (but this is my personal opinion) I think that the neural network are quite outworn (at least for the application you mentioned) because they have been surpassed by more performant methods derived by statistic learning (SVM family first of all).