Entity Linking

Step1 : Graph Embedding using the DistMult model (https://www.microsoft.com/en-us/research/wp-content/uploads/2017/05/emnlp15.pdf)

Setup

1. Using the name of the database (<data_name>) create a folder in $PROJECT_ROOT/data. e.g. $PROJECT_ROOT/data/umls or $PROJECT_ROOT/data/fb15k-237
2. In the above create another folder raw_text and copy all the following data resources here.
   • train.txt
   • valid.txt
   • test.txt
3. The knowledge data is in the format : entity1 \t entity2 \t relation \t 1
4. Create a conf file with <data_name>.conf in the folder $PROJECT_ROOT/config and update the parameters as given in config/umls.conf

Processing Data

./scripts/preprocess.sh <data.conf> graph

Training Models

To train a model use the run script with a data config and a model config like this:

IF running on CPU : ./scripts/train.sh configs/umls.conf config/dist_mult.conf graph

IF running on GPU : ./scripts/train.sh configs/umls.conf config/dist_mult.conf graph use_gpu

Step2 : Mention Context Embedding (http://cogcomp.org/page/publication_view/817)

Setup

1. Using the name of the database (<data_name>) create a folder in $PROJECT_ROOT/data. e.g. $PROJECT_ROOT/data/umls or $PROJECT_ROOT/data/ncbi_disease_corpus
2. In the above create another folder raw_text and copy all the following data resources here.
3. The data is in the PUBTATOR format
4. Create a conf file with <data_name>.conf in the folder $PROJECT_ROOT/config and update the parameters as given in config/ncbi_disease_corpus.conf

Processing Data

./scripts/preprocess.sh <data.conf> mentions

Training Models

To train a model use the run script with a data config and a model config like this:

IF running on CPU : ./scripts/train.sh config/ncbi_disease_corpus.conf config/joint_context.conf mentions

IF running on GPU : ./scripts/train.sh config/ncbi_disease_corpus.conf config/joint_context.conf mentions use_gpu