text_classification-of-biomedicine

Introduction

• The title of the paper is Biomedical document triage using hierarchical attention-based capsule network.
• It is my graduation project, mainly about text classification.
• It uses 3 data sets，including PM,IAS and ACT.More detailed information can be viewed in the paper.
• It's main innovation are hierarchical attention and capsule network.
• word2vec_PM is about datasets PM,word2vec_ISA is about ISA,word2vec_BC3_ACT is about ACT

Requirement

• python 3.5
• keras
• tensorflow 1.13.1
• sklearn
• gensim
• numpy

Usage

• 1. download word2vec vector. word2vec vector available:链接：https://pan.baidu.com/s/1Z3qsclRmcPNijubAkcLcWw 提取码：kvno
• 2. copy the folder vector to folder word2vec_PM,word2vec_ISA,word2vec_BC3_ACT.
• 3. run python word2vec_cnn_hierarchy_att.py
• Capsule.py:it is model of Capsule Network.
• attention.py:it is self-attention and hierarchical self-attention.
• data_helepers.py and data_helpers_att.py:Data preprocess
• w2v.py:word2vec vector
• word2vec_cnn_att.py and word2vec_cnn_hierarchy_att.py:train and test
• visualization.py:visualization code