openalex-concept-tagging

This repository contains all of the code for getting the OpenAlex concept tagger up and running. Go into the model iteration directory (V1 or V2) to find a more detailed explanation of how to use this repository. To learn more about concepts in OpenAlex, check out the docs.

Model Iterations

• V1 (no longer used)
• V2 (no longer used)
• V3 (currently used)

Both a V1 and a V2 model were created but as of right now, the V3 model is being used in OpenAlex. Initially, abstract data was not available for the model so we went with a V1 model that only looked at paper titles and a few other features. Once paper abstract data became available, a V2 model was created and we saw a substantial increase in performance. In order to meet the needs for some of our users, a V3 model was created which used the same base tagging model that was developed for V2 but added additional logic for assigning parent concepts so that all concepts would have a path to the top of our concept tree/graph. For more information, please read the information at the top of the V3 directory.

Model Development

You can find an explanation of the modeling and deployment process at the following link: OpenAlex: End-to-End Process for Concept Tagging

Concepts

Input can be tagged with one or more of about 65,000 concepts, listed here. Concepts are part of a hierarchical tree, with levels 0 (e.g., Mathematics) through 5 (e.g., Generalized inverse Gaussian distribution).

Internal Usage

Run these following commands in the root directory of the repository to get the model up and running locally in a Docker Container.

Configure AWS

aws configure

Download model artifacts

aws s3 cp s3://openalex-concept-tagger-model-files/ . --recursive

Copy the files to the appropriate directory

mkdir /V3/003_Deploy/model_to_api/container/model

cp ./V3/tag_id_vocab.pkl ./V3/003_Deploy/model_to_api/container/model/tag_id_vocab.pkl
cp ./V3/topics_vocab.pkl ./V3/003_Deploy/model_to_api/container/model/topics_vocab.pkl
cp ./V3/ancestor_chains.pkl ./V3/003_Deploy/model_to_api/container/model/ancestor_chains.pkl
cp ./V3/doc_type_vocab.pkl ./V3/003_Deploy/model_to_api/container/model/doc_type_vocab.pkl
cp ./V3/journal_name_vocab.pkl ./V3/003_Deploy/model_to_api/container/model/journal_name_vocab.pkl
cp ./V3/level_0_1_ids.pkl ./V3/003_Deploy/model_to_api/container/model/level_0_1_ids.pkl
cp ./V3/paper_title_vocab.pkl ./V3/003_Deploy/model_to_api/container/model/paper_title_vocab.pkl

mkdir /V3/003_Deploy/model_to_api/container/model/mag_model_V3

cp ./V3/model ./V3/003_Deploy/model_to_api/container/model/mag_model_V3

Change the file path to the model in the predictor.py file

# replace mag_model_V2 with mag_model_V3
sed -i 's/mag_model_V2/mag_model_V3/g' ./V3/003_Deploy/model_to_api/container/mag_model/predictor.py

Build Docker Container

docker build ./V3/003_Deploy/model_to_api/container -t openalex-concept-tagger -f ./V3/003_Deploy/model_to_api/container/Dockerfile

Run Docker Container

docker run -p 8888:8080 openalex-concept-tagger

Env Variables

# The amount of workers that will be used to serve the model
MODEL_SERVER_WORKERS=1