How to train new model?

[snakecy]

As the tool supply the funcion of "NSFC Subject Classifier", which input is chinese. so how can i train the same model for english version? Q: A) What is the train data format? B) What is the definition of the level? C) What is the training tool?

so appreciation for your reply.

[wengenihaoshuai]

hi,let me answer your question. (A) We use abstracts of paper to train the model. The lable is NSFC uses a three-level depends on the classification of Natural Science Foundation of China(NSFC)
e.g: " A04 abstract_paper" ",when predic the level-1 subject. The predict for subject level-2, you should do like that "A040412 abstract_paper" .We trained three models, in order to predic Three Level Disciplines. (B) NSFC three-level depends on the classification of Natural Science Foundation of China(NSFC) (C) In this cases ,We use fasttext model as a classifier. You can also replace it with other model that suits your work

in fact，the answer for your Q2，Q3 is in the classifier.py lied on "prediction_api/src/classifier.py". see the line 4 and line 25.

[snakecy]

@wengenihaoshuai Do you mean the model trained for each level separatly?

[wengenihaoshuai]

Not one model, but three.We trained three models for three levels separatly. It is here that, the code at line25-27 in prediction_api/src/classifier.py is used for load three model.

[snakecy]

@wengenihaoshuai Okay, Thanks a lot!

[snakecy]

@wengenihaoshuai

We use abstracts of paper to train the model. The lable is NSFC uses a three-level depends on the classification of Natural Science Foundation of China(NSFC)
e.g: " A04 abstract_paper" ",when predic the level-1 subject. The predict for subject level-2, you should do like that "A040412 abstract_paper" .We trained three models, in order to predic Three Level Disciplines.

I download the AMiner paper data, which does not label subject category. How do I map the paper label to NSFC？