Abstract: In some implementations, the techniques described herein relate to a method including: receiving data from a plurality of heterogeneous data sources; processing each type of received data through a corresponding type-specific processing pipeline by performing temporal pattern extraction specific to each data type; generating cross-source feature vectors by performing temporal correlation analysis across the processed data from different data sources; training an ensemble machine learning model using the cross-source feature vectors; generating engagement predictions and feature contribution values that quantify an impact of each feature on each prediction; presenting the engagement predictions with natural language explanations based on the feature contribution values; and updating weights of the ensemble machine learning model based on user feedback regarding the natural language explanations.