I have a question on NER? Suppose I have set of documents and do manual tagging for all the names in that document. Now based on those tagged words I train my model (using CRF), i.e. add the tagged words to the feature set. Note here by name i mean either one of the following - person name or medicine name or disease name

Now my question is , if i want to add more tagged data to train model , how will following approach effect model generation 1) I follow the same process, i.e. tag the names in documents and add new document to list of already present documents. 2) instead of tagging i manually extract the names and skip POS tagging step and directly add the names to feature set.

My assumption is 2) will yield bad model because CRF is probabilistic model and hence context is necessary. Thoughts??

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.