BioASQ-10 task b and synergy

Deep Learning solutions based on fixed contextualized embeddings from PubMedBERT on BioASQ 10b and traditional IR in Synergy

This repository exposes, in a reproducible way, our submission code for the 10th edition of the BioASQ. Given that we enrolled in task b and synergy, we organized the code separately for both challenges in the taskb and synergy folders respectively.

Installation with docker (easy approach)

TODO, will be available when everything is configured inside a docker image

Standard installation

First, make sure you have at least python3.8 installed. (We used python 3.8.10). Then, we recommend creating a virtual environment using the venv package, so if not installed please install the package python3.8-venv.

Next, create a virtual python environment and install the requirements needed for this repository. For instance, the code below creates, activates and updates an environment called "py-bioasq"

$ python3.8 -m venv py-bioasq
$ source py-bioasq/bin/activate
$ python -m pip install --upgrade pip
$ pip install -r requirements.txt

After this step, run the setup.sh script in order to download the reminder larger files that are required for this repository. Note that this step will probably take a long time since it will download the collections checkpoints, models checkpoints, cache files, datasets and testsets.

$ ./setup.sh

At this point, everything is set up and ready to run. To reproduce our submission runs for both tasks just execute run.sh. Furthermore, what this script does is just executing the run.sh script present in the synergy and taskb folders. So, if you are just interested in one of the task, just go to the respective folder and execute the local run.sh script.

$ ./run.sh

Directory structure

.
├── synergy/
│       Code base related to the synergy challenge.
│
└── taskb/
        Code base related to the taskb challenge (document retrieval and yes or no answering).

System specifications

For computational reference, our experiments were performed on a server machine with the following characteristics:

• Operating system: Ubuntu 18.04
• CPU: Intel Xeon E5-2630 v4 (40) @ 3.1GHz
• GPU: NVIDIA Tesla K80
• 128 GB RAM

Team

• Tiago Almeida¹
• André Pinho¹
• Rodrigo Pereira¹
• Sérgio Matos¹

1. University of Aveiro, Department of Electronics, Telecommunications and Informatics (DETI), Institute of Electronics and Informatics Engineering of Aveiro (IEETA), Aveiro, Portugal ↩

Reference

Please cite our paper if you use this code in your work:

TODO: Fix this reference when the paper becomes available.

@article{almeida2022a,
  author    = {Almeida, Tiago and Pinho, André and Pereira, Rodrigo and Matos, S{\'e}rgio},
  title     = {Deep Learning solutions based on fixed contextualized embeddings from PubMedBERT on BioASQ 10b and traditional IR in Synergy},
  url       = {https://github.com/bioinformatics-ua/BioASQ-10},
  volume    = {2022},
  year      = {2022},
}