Ad Quality Prediction via Session Feature Segmentation

The patent segments the single CTR metric into multiple distinct features including click-through rate, impression count, time of day, day of week, geographic location, and user demographics. Each feature is measured and stored separately in a data structure, allowing for more nuanced quality assessment while maintaining operational simplicity through automated collection of each segment.

The patent transitions from a one-dimensional CTR measurement to a multi-dimensional quality assessment by adding temporal dimensions (time of day, day of week), spatial dimensions (geographic location), and user-specific dimensions (demographics, browsing history). This dimensional expansion enables precise quality measurement while the systematic data structure maintains ease of operation.

The patent performs preliminary actions by pre-collecting and storing all necessary advertisement features, user behaviors, and contextual data in a structured data structure before quality prediction is needed. Historical data including past performance metrics and user interactions are accumulated in advance, enabling accurate real-time predictions without complex processing during the prediction moment itself.

The patent introduces an intermediary data structure that acts as a mediator between raw data collection and quality prediction. This structured format organizes diverse data elements (CTR, impressions, temporal features, geographic data) into a unified schema, simplifying the overall system complexity while enabling comprehensive analysis for accurate predictions.

The patent applies partial action by selecting and applying only the most relevant features and statistical models for each specific quality estimation task rather than processing all possible data elements. The system uses targeted feature subsets and appropriate statistical methods based on the specific advertisement and context, reducing computational energy while maintaining high estimation accuracy.

The patent changes parameters by dynamically adjusting which features are considered and how statistical models are applied based on the specific advertisement characteristics and context. The system modifies computational parameters such as feature importance weights and model selection criteria to optimize the balance between accuracy and computational energy consumption for each prediction scenario.