extract-keywords-from-youtube-videos

This project combines youtube-dl, whisper, LangChain and ChatGPT to extract keywords from YouTube videos. It was intented as a tool for Lyon Data Science to better reference its videos.

Setup

To build the docker image, run the following command:

docker build -t extract-keywords-from-youtube-videos .

Now you can run a container like this:

docker run -it --gpus all extract-keywords-from-youtube-videos

Note: It is better to have a GPU to run this project as whisper will run faster. I suppose it could work on CPU but it would be very slow and I did not test it.

Usage

Set OpenAI API Key

As ChatGPT is used, you need to enter a valid OpenAI API key. You can get one here. Then, run the following command:

cp .env.template .env

And write your API key in the .env file.

Run the pipeline

This project uses DVC to orchestrate the pipeline. At any point, run dvc status to see the status of the pipeline.

You can see and edit the list of videos to process in dvc.yaml. For example:

vars:
  - videos:
    - name: Où va la Data Science ?
      url: https://www.youtube.com/watch?v=bONAu-W44b8
    - name: Techniques pour améliorer son éligibilité pour travailler dans la data
      url: https://www.youtube.com/watch?v=N-7zRvzRr0I
    - name: Le long parcours d'un cas d'usage applicatif en NLP, de 2013 à 2022
      url: https://www.youtube.com/watch?v=Ki0pKl3TwdU

To run the pipeline, run dvc repro.

Warning: Be careful when running the pipeline as it will use ChatGPT which will cost you money. When possible, try to retrieve the results from the cache using dvc pull and dvc checkout.

Note: For now there is no DVC remote which means that your cache is only stored locally and there is no way to dvc push/pull it to synchronize with others. However you can already look at the results in the data folder. In the future, I will add a DVC remote on Google Drive to store the cache, in particular the audio files extracted from the videos (which would be too heavy to store with Git).