clinical-trials-search-engine

Clinical Trials Search Engine for Information Retrieval and Recommender System Project

---

Disclaimer

This project has been developed for academical purposes only.

This project is released under the GNU General Public Licence.

---

Setup

Clone the repository locally:

• Git : git clone https://github.com/MikiTwenty/clinical-trials-search-engine
• GitHub : gh repo clone MikiTwenty/clinical-trials-search-engine

Requirements

• Virtual Environment with Python >= 3.10
• Required Python packeges (see requirements.txt).
• Java SDK

Download Clinical Bert Model

Get files from Hugging Face:

• git lfs install
• git clone https://huggingface.co/medicalai/ClinicalBERT

Download LLM GGUF Model files

You can find the models on Hugging Face. Suggested models are models with 7B parameters such as:

• Llama 2 Chat 7B GGUF
• Mistral Instruct 7B v0.2 GGUF

Enable GPU computing (Optional)

• PyTorch : see documentation
• llama-cpp-python : see documentation

---

Usage

Initialize the Graphical User Interface to run the SE.

src.gui.GUI

Initialize the Graphical User Interface (powered by NiceGUI).

Parameters

• paths (Dict[str, str]): The paths dictionary.

  The paths keys must be:

  'DATA': the folder containing all the data;

  'DATASET': the dataset.pkl file;

  'DOCUMENTS': the folder containing all the dataset files;

  'INDEXING_FILES': the folder for the indexing files;

  'LLM': the folder containg the LLM GGUF files;

  'BERT': the folder containg ClinicalBert model files;

  'JDK': the folder containing Java SDK;

  'EVAL': the folder containing the evaluation files (optional);

  'LOG': the log.txt file (optional).

• verbose (bool): If True, enables verbose output.

GUI.start()

Initialize the SE and start the GUI.

Example

• To use the SE run.py

• To preprocess the data train.py

---

Info

• Univeristy of Pavia
• Artificial Intelligence BSc
• Information Retrieval and Recommender Systems
• Authors: Michele Ventimiglia (@MikiTwenty), Manuel Dellabona (@manudella)