Deep Neural Network User Segmentation via Feature Embedding

The system segments users into distinct cohorts based on their behavior patterns, transforming complex user data into manageable groups. This segmentation approach allows the deep neural network to process user features more effectively by dividing the large-scale user base into smaller, analyzable segments, thereby improving segmentation accuracy while managing system complexity through structured data organization.

The patent introduces an intermediary processing layer that transforms raw user behavior data into meaningful features before feeding them to the deep neural network. This intermediary step includes feature extraction, selection, and engineering processes that simplify the input data structure, reducing the complexity burden on the neural network while maintaining or enhancing segmentation accuracy.

The system performs preliminary actions by pre-processing and engineering user behavior features before the main analysis. This includes creating aggregated metrics, transforming raw data into meaningful indicators, and preparing feature sets in advance. By performing these preparatory steps beforehand, the system reduces the processing time required during actual prediction while maintaining comprehensive analysis of historical user features for accurate predictions.

The patent extracts only the most relevant and informative features from the complex historical user data, rather than processing all available data. This feature extraction process identifies and isolates key behavioral indicators that drive prediction accuracy, eliminating redundant or less significant features. By taking out only the essential features, the system achieves high prediction accuracy with reduced processing time.