arxiv-classifier

How to Train

nb = ArticleClassifier(dbpath='/path/to/db')
fn = nb.create_input_file(metadata)
nb.train(fn)
nb.save()

How to Classify

nb = ArticleClassifier(dbpath='/path/to/db')
nb.load()
fn = nb.create_input_file(metadata)
classes = nb.classify(fn)

In these examples, metadata should be a List of Dict where the Dict are in the format given below. The file name paths are relative to the current machine (on the local file system):

{
  "id": "1704.00222",
  "categories": ["cs.NM", "hep-th"],
  "filename": "/path/to/file.txt"
}

Trained Models

Trained model for use by arXiv staff can be found at s3://arxiv-classifier-models

ULMFiT classifier

Training

See experiments directory for training and evaluation notebooks.

Models

The ULMFiT and SentencePiece model files can be downloaded here. Make sure CLASSIFIER_PATH configuration parameter points to models/abstracts-classifier.pkl and that CLASSIFIER_TYPE equals ulmfit.

Testing

To test the service locally you can run it with

FLASK_APP=classifier.test_app flask run --port 9999

and make a request:

curl -s -H "Content-Type: application/json" -X POST http://localhost:9999/classify \
    --data '{"title":"P = NP", "abstract": "We prove that P = NP for N = 1 or P = 0.", "primary": "cs.SE"}'

[{"category":"cs.CC","probability":0.8264293074607849},{"category":"cs.DS","probability":0.1285623162984848},...]

The primary is optional.

Both the input and output format are not yet compatible with the Naive Bayes classifier.

fschwenn/arxiv-classifier