Disease Normalization with Graph Embeddings

Table of Contents:

1. Basic training and Testing
2. Structure of code
3. Paper

Basic Usage:

Please view the usage in train_NER_script.py, train_EL.py, train_MTL_script.py and test_NER.py, test_EL.py, test_MTL.py respectively for training and testing.

Paths and Params are two object which needs to be initialized with appropriate values.

Structure of Code:

Folder descriptions:

• analysis: Execute analyze_data2.py and analyze_mesh.py after updating the data paths.
• config: definition of Paths and Params objects
• EL: files containing the EL training and testing methods. EL_partA.py is used to train node2vec type-I and type-II. EL_GCN.py is used to train GCN. EL_utils.py contains functions used in common for EL. Rest are experimental files.
• models: definition of NER and EL models. Other than summery.py, the files define models specific to the task. summery.py is experimental.
• MTL: file containing training and test functions for Multitask Learning
• NER: file containing training and test functions for Named Entity Recognition using NERDS
• nerds: NERDS (contains updated NER models)
• node2vec: Contains files of basic node2vec algorithm (main.py and node2vec.py ) as well as unsupervised training files of the two variant Type-I (node2vec1.py) and Type-II (node2vec2.py). SkipGram.py and Another_sk.py are experimental files.
• process: utility functions for preprocessing MeSH data
• utils: Utility functions (with extra/old codes)
• wvlib_master: Package used to read pre-trained PubMed word embedding.

Paper:

• Here is the arxiv paper: Link

When referring to this work please cite:

• Dhruba Pujary, Camilo Thorne and Wilker Aziz, Disease Normalization with Graph Embeddings. 2020, Proceedings of the IntelliSys 2020 conference (preprint available at: http://camilothorne.altervista.org/perso/intellisys-2020.pdf).