Consumer Health Question Answering Using Off-the-shelf Components

Source code and files for the Consumer Health Question Answering Using Off-the-shelf Components [Paper] accepted at the ECIR 2023 Conference

• data:

  • raw xml data (search_topics folder)
  • 113 test questions (data_to_process.csv)
  • results from retrieval stage

• retrieval: scripts for pipeline's retrieval stage (Evidence Search)

• experiments: notebooks for QA models inference, scores aggregation and metrics calculation (Question Answering)

• finetune: notebook for fine-tuning Roberta Large pre-trained model on BoolQ dataset

• statistics: notebooks for statistics calculation (see Table 2, columns "Hits" and "#0")

The pre-trained BioLinkBERT models used for expreiments can be found here.

Citation:

@InProceedings{pugachev:2023,
  address =                  {Berlin Heidelberg New York},
  author =                   {Alexander Pugachev and Ekaterina Artemova and Alexander Bondarenko and Pavel Braslavski},
  booktitle =                {Advances in Information Retrieval. 45th European Conference on IR Research (ECIR 2023)},
  month =                    apr,
  publisher =                {Springer},
  series =                   {Lecture Notes in Computer Science},
  site =                     {Dublin, Ireland},
  title =                    {{Consumer Health Question Answering Using Off-the-shelf Components}},
  year =                     2023
}