This DataFrame was created to serve as the comprehensive dataset for training a supervised machine learning model in KNIME to predict user interaction scores.
Define Target Variable (interaction_score):
Source: df_interactions
Process: The interaction_type column ('play', 'like', 'skip') was converted into a numerical interaction_score. We used a binary classification approach: 'like' and 'play' were mapped to 1 (positive engagement), and 'skip' was mapped to 0 (negative engagement).
Prepare Track Features (df_track_features):
Source: df_synthetic_tracks
Process: Numerical track features (e.g., duration_ms, popularity, danceability, energy) were scaled using MinMaxScaler. Categorical track features (e.g., genre, country, explicit, label) were transformed using OneHotEncoder.
Result: A new DataFrame, df_track_features, containing only scaled numerical and one-hot encoded track features, with track_id as the index.
Combine All Data (df_modeling_data):
Source: df_interactions (with interaction_score), df_track_features, df_users.
Process: These three DataFrames were merged using user_id and track_id as keys. This step brings together user information, track features, and the interaction target into a single table.
One-Hot Encode User Features (within df_knime_ready):
Source: df_modeling_data
Process: The remaining categorical user features (gender, location, preferred_genres) were one-hot encoded (pd.get_dummies).
Final df_knime_ready Creation:
Result: The dataset was consolidated into df_knime_ready, containing user_id, track_id, all preprocessed track and user features (now entirely numerical), and the interaction_score target variable. This DataFrame is now fully numerical and ready for direct consumption by most machine learning nodes in KNIME.
Double-click dialogue box to scroll and view
To use this workflow in KNIME, download it from the below URL and open it in KNIME:
Deploy, schedule, execute, and monitor your KNIME workflows locally, in the cloud or on-premises – with our brand new NodePit Runner.
