Given a collection of documents - each corresponding to some economic entity - I am looking to extract information and populate a table with predetermined headings. I have a small sample of this already done by humans and I was wondering if there's an efficient way to automatise it. Grateful for any suggestions.
1 Answer
Caution: it depends on what the data looks like, but in general this can be complex and it might not work well.
The general task is sequence labeling: the system is supposed to annotate a text and identify specific spans of text as the target categories. The most standard task is Named Entity Recognition (NER), for which there are many libraries available. You would have to train a custom model using your labelled data (in the specific format expected by the system). In general a model needs a fair amount of example to capture the patterns properly, it's not sure at all that this can work with only a small amount of data.