#stanford-ner

Train, evaluate, run, and annotate using the Stanford NER tool in Scala.

##Annotation Use the InteractiveTraining command line annotation app for named entity annotation. During annotation, the results are constantly flushed to a text file that records all annotations in a tab-separated format that can the be read by the NER tool to train a model.

##Training Given some training data (possible obtained by using the above annotation tool), you can train a new NER model using CrfTrainer. For example,

CrfTrainer.trainClassifier("training/data/file.txt","model/save/location.ser.gz")

Where the first argument is your training data location and the second argument is where you want to save your trained model.

##Running So you have trained your model and now you want to run it on some text and extract the named entities within. This can be done with EntityExtraction. For example,

val entities=EntityExtraction.extract(myNerModel, myText)

where the first argument is your trained ner model and the second argument is the text. The result is a Seq[SentenceEntityTokens] which splits myText into sentences. Each sentence is a sequence of tokens and corresponding predicted entity types.

##Evaluating So you have your training data but you want to know how well the resulting NER model will perform in general. In order to do this you can perform k-fold cross-validation using CrossValidation. For example,

val xValResults=CrossValidation.runXVal(numFolds, "training/data/file.txt", "fold/write/location")

Where numFolds is the number of cross-validation folds, "training/data/file.txt" is the training data file, and "fold/write/location" is the directory where the training and validation sets will be written. Note: This only works if for the case of 2 entity types right now.