This University Project aims to provide a Framework which can be used to analyse a commercial and identify whether or not this commercial contains a Brand Differentiating Message (BDM). This Spreadsheet contains a list of Superbowl Ads which are labeled with either 1 (BDM) or 0 (No BDM) by the marketing faculty of my University. Due to limitations in Data Quality (Subjectivity, Binary Classification) and Quantity (only a few hundred ads, class imbalance), this project should be considered as more of a conceptual approach, rather than a model with a high accuracy.
Due to the small number of Ads, we used a machine learning approach, with manually engineered features, rather than just passing text through to a neural network.
This Project is structured using the CRISP-DM Framework
video_tutorial.mp4
- Docker
- docker-compose
- (Model Training Only) A URL with the Videos referenced in the model training. This must be added to your local .env file. Copy the example and adjust the URL
The model was developed on the following System
| Component | Details |
|---|---|
| OS | Ubuntu 24.04.1 LTS x86_64 |
| Host | MS-7E07 1.0 |
| CPU | 13th Gen Intel i7-13700K (24) @ 5.500GHz |
| GPU | NVIDIA GeForce RTX 4070 |
| Memory | 64042 MiB |
In in attempt to ensure cross plattform compatibility, a Docker image was created. However this has only been tested on 1 machine and needs to be tested on a broad spectrum of systems to ensure true Cross Plattform Compatibility
To install, simply run the following command:
docker-compose up -dOpen http://localhost:8502 to access the Demo Web UI. This will allow you to upload a Video and see the steps that our Model goes through before giving its prediction whether or not the Video contains a strong BDM or not.
The source code is directly mapped to the app directory. This means that any changes made to the directory inside or outside of the container will be synced both ways.
Open http://localhost:8889 to access the integrated Jupyter Development environment. The necessary dependencies are already installed.
In order to authenticate and set a password, in your terminal run the following commands:
docker exec -it app sh
jupyter server listThis will retrieve the token.
If you want to persist the changes, make sure to update the requirements.txt when it comes to python packages and any external dependencies must be added to the Dockerfile.