Extraction of specific expressions

[gelinger777]

I am very new to NER and AI , so I want to apologize if the question is very "newbie-like".
I have a very specific text extraction need which I think AI can easily handle if trained in a right way.
Let's say we have many court case desicions . Take as an example this one

https://caselaw.findlaw.com/us-supreme-court/17-5554.html

For example lets see first paragraph

Petitioner Stokeling pleaded guilty to possessing a firearm and ammunition after having been convicted of a felony, in violation of 18 U. S. C. §922(g)(1). Based on Stokeling's prior criminal history, the probation office recommended the mandatory minimum 15-year prison term that the Armed Career Criminal Act (ACCA) provides for §922(g) violators who have three previous convictions "for a violent felony," §924(e). As relevant here, Stokeling objected that his prior Florida robbery conviction was not a "violent felony," which ACCA defines, in relevant part, as "any crime punishable by imprisonment for a term exceeding one year" that "has as an element the use, attempted use, or threatened use of physical force against the person of another," §924(e)(2)(B)(i). The District Court held that Stokeling's actions during the robbery did not justify an ACCA sentence enhancement, but the Eleventh Circuit reversed.

I want to train my Neural Network in a way that when I feed in this text in the outcome will be an array with

["ACCA"=>[' 922',' 924],"18US"=>['922'] ]

Is it possible with Sequence Tagging? Or we should look into other direction?

[guillaumegenthial]

Hi @gelinger777 ,
Sorry for the delay, this is an interesting question. I would be interested in learning what was your final approach.
I think it would be feasible with sequence tagging: you would find mentions in the text of law articles, mentions etc. and then would apply some heuristics (or some relation prediction, maybe as simple as chunking based heuristics) to map them together.

[gelinger777]

@guillaumegenthial we have trained Whatson AI from IBM to do the work actually. It was an interesting experience. We could achieve that whatson is given the text and it gives back the json with relations and entities.