We are proud to have launched our first competition on Kaggle!
The primary challenge of this competition is to develop an AI tool or method that can accurately extract questionnaire questions from documents, primarily PDFs.
This competition offers a unique opportunity for participants to contribute to the field of natural language processing and document analysis as well as open source for social science while developing solutions that have real-world applications. We encourage participants to think creatively, leverage available resources, and push the boundaries of current technologies.
Requirements: Python 3.10 or greater
Create an account on Kaggle.
Install Kaggle on your computer:
pip install kaggle
On the Kaggle website, download your kaggle.json
file and put it in your home folder under .kaggle/kaggle.json
.
Download and unzip the competition data:
kaggle competitions download -c harmony-pdf-and-word-questionnaires-extract
unzip harmony-pdf-and-word-questionnaires-extract.zip
To generate predictions for the training data and write to train_predictions.csv:
python create_sample_submission.py train
To evaluate the train predictions:
python evaluate_train_results.py
To modify the prediction logic or inject your own model, you can edit the function dummy_extract_questions
.
To generate predictions for the test data and write to submission.csv:
python create_sample_submission.py test
kaggle competitions submit -c harmony-pdf-and-word-questionnaires-extract -f submission.csv -m "Message"
[Beta mode: we are currently testing this extension] We have developed a browser extension for Harmony called “Send to Harmony” which lets you send selected text to Harmony with a right-click. For PDFs, use the popup to paste your selected text. Send to Harmony enables users to send selected text to the Harmony Data Harmonization (https://harmonydata.ac.uk/) platform for analysis. This plugin provides a right-click or context menu item which allows users to easily bring text from into their harmonisations, making it easier to compare and analyze different measurement scales across research studies.
We have a number of exciting updates to Harmony including: some improvements to the R library which have been asked for by researchers around the world who have been using Harmony on studies in lots of different topics as well as making our own fine tuned large language model available in the web UI, which is José’s winning model from the DOXA challenge which ended on 10 January 2025. Harmony has its own Large Language Model!