Data Science Stack Exchange is a question and answer site for Data science professionals, Machine Learning specialists, and those interested in learning more about the field. It only takes a minute to sign up.
I would like train a classifier to spot news articles which would spot articles that announce an event. The issue is that I do not have a large pre-labeled dataset for this task (I only have 200 examples). So here are my two questions:
Have you heard of a training set labeled for such task
I heard of few shot learning that can be helpful to train a classifier with few example would it be applicable to this case and is there any library/reading you would recommend.
The general semi-supervised setting consists in training a model from an initially small training set by applying it iteratively to unlabelled instances. There are various methods to minimize the risk of training the model on wrongly classified instances.
Active learning is a variant of semi-supervised learning where the model queries the human expert for annotations, but the instances are carefully selected in order to minimize the amount of human labor.
There is also bootstrapping, where one would focus on the positive instances: apply the original model to the unlabelled data, the manually annotate only the instances which are predicted as positive (useful only in cases where the positive class is much smaller than the negative one).