The code for DASFAA(2023) paper Semi-Supervised Learning for Fine-grained Entity Typing with Mixed Label Smoothing and Pseudo Labeling
- numpy==1.15.4
- tensorflow==1.14.0
- pandas==0.23.4
- gensim==3.8.3
- scikit_learn==0.23.1
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
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.
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.