Inspiration
We wanted a meaningful challenge with regards to optimising and creating more efficient ways to allocate high priority, albeit problematic or high value, advertisements to appropriate moderators with suitable performance scores.
What it does
Our model generates a Priority score and Performance score through the advertisements and moderators' factors respectively. From this, we are able to match the Priority scores to Performance scores after segmenting them to the countries and subregions that both advertisements and moderators belong to.
How we built it
We first did Exploratory Data Analysis through Seaborn, followed by Sci-kit Learn to scale and transform the data during pre-processing. Finally, we used stochastic multi-objective optimisation through Gurobipy to create the final scores for both moderators and advertisements. We then used our customised Mapping algorithm to match the moderators to ads.
Challenges we ran into
- Incomplete data
- Lack of labelled features for data which led to unsupervised optimisation approach
- Unable to apply other algorithms such as Pareto Front-based or Munkres algorithm for matching
Accomplishments that we're proud of
- Finished using Agile methodology project management, mainly sprints
- Learning and successfully applying stochastic multi-objective optimisation
- Relatively acceptable final outputs
What we learned
- Stochastic gradient descent using Gurobipy
- Feature engineering
- Utilising external dataset to improve results
What's next for TikTak Eaters
- Internship at TikTok please
Built With
- gurobipy
- pandas
- python
- scikit-learn
- seaborn
Log in or sign up for Devpost to join the conversation.