ISGuXing / NFETC-SSL

NFETC-SSL

The code for DASFAA(2023) paper Semi-Supervised Learning for Fine-grained Entity Typing with Mixed Label Smoothing and Pseudo Labeling

Request

• numpy==1.15.4
• tensorflow==1.14.0
• pandas==0.23.4
• gensim==3.8.3
• scikit_learn==0.23.1

Data

For generate wikim(wiki with the improved hierarchy) raw data, obtained from NFETC.
For generate OntoNotes and BBN raw data, download data in data directory using download.sh. Run command: sed -i "1i 2196017 300" data/glove.840B.300d.txt to insert a line "2196017 300" to the head of data/glove.840.300d.txt. You also can be obtained from NFETC-CLSC

Train and Evaluation

Run python eval.py -m <model_name> -d <data_name> -r <runs> -g <gpu> and the scores for each run and the average scores are recorded in one log file stored in folder log.

• <data_name> choices: wikim, ontonotes, bbn
• <model_name> choices: wikim, ontonotes, bbn
• <runs>: the number of repetitions of the experiment.

Acknowledge

Code is based on previous work: NFETC-AR, NFETC-CLSC and NFETC, many thanks to them.
For more implementation details, please read the source code.