I'm trying to write a complex spell checker for proper name. I must implement this as machine learning...
I thought that the best solution should be use bayes classifier which are implemented into weka library.
Now I have the problem how I should train the machine. I'd like to have so confusion matrix on which I can find how often occur a specific event.
I explain myself better, we can assume that a mispelled word can be classifier in one of these class :
- SUBSTITUTION when a letter is substitude with another
- TRANSPOSITION when you have to swap two letter
- INSERTION when you have to insert a letter
- DELETION when you delete a letter
For each of these classes I'd like to have confusion matrix in which I can find how common is make a mistake with those letters.
I need this to elaborate my Error model.
I've a limited knowledge in Machine learning and AI.. Is this the right way to address this problem? If so how can I structure the training data for weka library?
I've already read the Weka Docs, but I can't figure out how I should structure the training set. I'd like just to have as training data a list of common misspelled word and as model language a list of most common name.
Have I been clear explaining my problem?