Position-aware self-attention with three positional biases improves Bi-LSTM-CRF by 0.08 to 0.32 percent on POS tagging, NER, and chunking, but possible test-set hyperparameter selection undermines the result.
Empower sequence labeling with task-aware neural language model,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2019 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Position-Aware Self-Attention based Neural Sequence Labeling
Position-aware self-attention with three positional biases improves Bi-LSTM-CRF by 0.08 to 0.32 percent on POS tagging, NER, and chunking, but possible test-set hyperparameter selection undermines the result.