deep-keyphrase

Implement some keyphrase generation algorithm

Description

Implemented Paper

CopyRNN

Deep Keyphrase Generation (Meng et al., 2017)

ToDo List

CopyCNN

CopyTransformer

Usage

required files (4 files in total)

1. vocab_file: word line by line (don't with index!!!!)

   this
   paper
   proposes
   

2. training, valid and test file

data format for training, valid and test

json line format, every line is a dict:

{'tokens': ['this', 'paper', 'proposes', 'using', 'virtual', 'reality', 'to', 'enhance', 'the', 'perception', 'of', 'actions', 'by', 'distant', 'users', 'on', 'a', 'shared', 'application', '.', 'here', ',', 'distance', 'may', 'refer', 'either', 'to', 'space', '(', 'e.g.', 'in', 'a', 'remote', 'synchronous', 'collaboration', ')', 'or', 'time', '(', 'e.g.', 'during', 'playback', 'of', 'recorded', 'actions', ')', '.', 'our', 'approach', 'consists', 'in', 'immersing', 'the', 'application', 'in', 'a', 'virtual', 'inhabited', '3d', 'space', 'and', 'mimicking', 'user', 'actions', 'by', 'animating', 'avatars', '.', 'we', 'illustrate', 'this', 'approach', 'with', 'two', 'applications', ',', 'the', 'one', 'for', 'remote', 'collaboration', 'on', 'a', 'shared', 'application', 'and', 'the', 'other', 'to', 'playback', 'recorded', 'sequences', 'of', 'user', 'actions', '.', 'we', 'suggest', 'this', 'could', 'be', 'a', 'low', 'cost', 'enhancement', 'for', 'telepresence', '.'] ,
'keyphrases': [['telepresence'], ['animation'], ['avatars'], ['application', 'sharing'], ['collaborative', 'virtual', 'environments']]}

Training

download the kp20k

mkdir data
mkdir data/raw
mkdir data/raw/kp20k_new
# !! please unzip kp20k data put the files into above folder manually
python -m nltk.downloader punkt
bash scripts/prepare_kp20k.sh
bash scripts/train_copyrnn_kp20k.sh

# start tensorboard
# enter the experiment result dir, suffix is time that experiment starts
cd data/kp20k/copyrnn_kp20k_basic-20191212-080000
# start tensorboard services
tenosrboard --bind_all --logdir logs --port 6006

Notes

1. compared with the original seq2seq-keyphrase-pytorch

   1. fix the implementation error:

      1. copy mechanism
      2. train and inference are not correspond (training doesn't have input feeding and inference has input feeding)

   2. easy data preparing
   3. tensorboard support
   4. faster beam search (6x faster used cpu and more than 10x faster used gpu)