README.md

metadata

language: ar
license: apache-2.0
datasets:
  - AQMAR
  - ANERcorp
embeddings:
  - GloVe
  - Flair

Arabic NER Model using Flair Embeddings

Training was conducted over 94 epochs, using a linear decaying learning rate of 2e-05, starting from 0.225 and a batch size of 32 with GloVe and Flair forward and backward embeddings.

Results:

• F1-score (micro) 0.8666
• F1-score (macro) 0.8488

	tp	fp	fn	precision	recall	class-F1
LOC	539	51	68	0.9136	0.8880	0.9006
MISC	408	57	89	0.8774	0.8209	0.8482
ORG	167	43	64	0.7952	0.7229	0.7574
PER	501	65	60	0.8852	0.8930	0.8891

---

2020-10-27 12:05:47,801 Model: "SequenceTagger(
  (embeddings): StackedEmbeddings(
    (list_embedding_0): WordEmbeddings('glove')
    (list_embedding_1): FlairEmbeddings(
      (lm): LanguageModel(
        (drop): Dropout(p=0.1, inplace=False)
        (encoder): Embedding(7125, 100)
        (rnn): LSTM(100, 2048)
        (decoder): Linear(in_features=2048, out_features=7125, bias=True)
      )
    )
    (list_embedding_2): FlairEmbeddings(
      (lm): LanguageModel(
        (drop): Dropout(p=0.1, inplace=False)
        (encoder): Embedding(7125, 100)
        (rnn): LSTM(100, 2048)
        (decoder): Linear(in_features=2048, out_features=7125, bias=True)
      )
    )
  )
  (word_dropout): WordDropout(p=0.05)
  (locked_dropout): LockedDropout(p=0.5)
  (embedding2nn): Linear(in_features=4196, out_features=4196, bias=True)
  (rnn): LSTM(4196, 256, batch_first=True, bidirectional=True)
  (linear): Linear(in_features=512, out_features=15, bias=True)
  (beta): 1.0
  (weights): None
  (weight_tensor) None