NPCRF

Neural Pairwise Conditional Random Field for Ultra-Fine Entity Typing

The NPCRF module proposed in the EMNLP 2022 paper: Modeling Label Correlations for Ultra-Fine Entity Typing with Neural Pairwise Conditional Random Field. NPCRF performs mean-field variational inference on a probabilistic model designed for better modeling label correlations in ultra-fine entity typing task.

Static Label Embeding

NPCRF requires static label embeddings, the preprocessed label embeddings (from GloVe for EN, Tencent for ZH) can be downloaded here: UFET, CFET, and you can place them in yoru folder and run the following config: (you need to reset your target_emb_dir in the config). Or you can provide the path of the glove embedding file (e.g., /path/to/your/glove.6B.300d.txt) and the code will generate label embedding for you.

Example of Training a model with NPCRF

python -m scripts.train -c examples/npcrf/configs/ufet_concat_npcrf.yaml

Benchmarks

UFET	ma-F1	ma-P	ma-R
NPCRF-roberta-large	47.1	49.5	44.9