For: DPG Media – Content blocker (Paywall) team
Project Goal
This project aimed to identify and evaluate combinations of rules that maximize user conversions and logins on DPG Media’s content blockers, while minimizing the loss of total views. By analyzing various content attributes and their impact on user engagement, I sought to create targeted strategies for content delivery and user interaction.
How it was built
The process began by loading two datasets—one with and one without a login wall filter—to calculate conversion and login rates. I then developed a system to generate all possible combinations of categorical features. For each combination, it calculated a combined metric score using normalized conversion and login rates. The core of the project is a ‘Knapsack’ algorithm, which efficiently selects rule combinations that maximize engagement while respecting predefined loss percentages for conversions and logins. The algorithm’s performance was evaluated by iterating through a range of these loss percentages, allowing us to discover the most efficient trade-offs between different engagement metrics. The results are visualized as 3D scatter plots, showing the impact of various parameters on the overall outcome.
Technologies used
- Python: Core programming language for data manipulation and analysis.
- Pandas: For handling and manipulating data in a structured format.
- NumPy: For efficient numerical computations and array operations.
- Scikit-learn: For data preprocessing (MinMaxScaler) and pipeline creation.
- Plotly: For creating interactive data visualizations.
- Pickle: For saving and loading analysis results.