This is the codes of the article ACESO: PICO-guided Evidence Summarization on Medical Literature, it includes all experimental codes and datasets.
- data clean
- data transform
- data train
- python 3.6
- pytorch 0.3.1
- visdom 0.1.8.5
- torchnet 0.0.4
- sklearn 0.20.1
- nltk 3.3.0
- gensim 3.4.0
- tqdm 4.28.1
- fire 0.1.3
- pandas 0.23.4
in the file of datasets/PICO/:
-
P.csv ~600
-
I.csv ~700
-
O.csv ~600
-
N.csv ~600
- start the visdom server : python -m visdom.server
- train: update the config.py and write the data location,then python main.py train
- test: update the config.py and then, python main.py test
please read the document about Visidom
| Model | Parameters | Value |
|---|---|---|
| CNN | dropout | 0.5 |
| CNN | kernel size | {2,3,4} |
| CNN | kernel number | 100 |
| CNN | epoch | 100 |
| CNN | initial learning rate | 0.01 |
| CNN | dimensions of embedding | 200 |
| Bi-LSTM | dropout | 0.5 |
| Bi-LSTM | epoch | 30 |
| Bi-LSTM | initial learning rate | 0.01 |
| Bi-LSTM | dimensions of embedding | 200 |
| Bi-LSTM | init | Orthogonal |
| Bi-LSTM | hidden size | 200 |
| Concept2Vec | diameter of hypercube | 5.50E-07 |
| Concept2Vec | dimensions of embedding | 108 |
| DeepWalk | number of sampled paths | 10 |
| DeepWalk | walk length | 40 |
| DeepWalk | windows size | 5 |
| DeepWalk | dimensions of embedding | 200 |
| Active Learning | wu,wd,wr | 1/1/1 |