Toy json data

[pierpaologoffredo]

Hello!
I am trying to use your model in order to perform the NER and RE task using for training and testing my annotation that I have to preprocess.

Looking at your toy example I think there's something missing inside the .jsonfile that are the entities.

Here is it your example:
[{"tokens": ["In", "1913", ",", "Olympic", "legend", "Jesse", "Owens", "was", "born", "in", "Danville", ",", "Ala."], "relations": [{"type": "Live_In", "head": 0, "tail": 1}], "orig_id": 5294, "tags": ["O", "O", "O", "O", "O", "B-Peop", "L-Peop", "O", "O", "O", "B-Loc", "I-Loc", "L-Loc"]}]

Here, the item "head" and "tail" should refer to the entities 0 (Jesse Owens) and 1 (Danville) if we look at the BILOU tags. Is the entity part of the json with all the entities necessary for the input data? Or am I wrong?

I'm having this issue firstly because I need to convert my annotation data to json format, and then also for the explanation of the possible types you put here.

I don't know if this is a stupid question, but since I would like to use your model for my experiments I need to adapt my data to yours.

Thank you in advance for your help and time.

Pierpaolo

[YoumiMa]

Hi Pierpaolo,

Many apologizes for the late reply ...
I believe there is no filed missing in the toy example. The pre-processing code automatically detects the index of each entity based on the BILOU tags. This makes the data a little bit harder to read, though.

I hope this is clear and answers your question.