Private Media Recommendations Without Profile Contamination

The patent segments the recommendation system into two independent models: a first recommendation model for the user's primary account that maintains recommendations based on primary interests, and a second recommendation model for the private interactive mode that handles temporary explorations. This segmentation allows each model to specialize in its specific function without interference, resolving the contradiction between system simplicity and recommendation accuracy.

The patent extracts the private interactive mode's recommendation logic from the main recommendation system. By separating the temporary exploration recommendations into a distinct second recommendation model, the system prevents recent explorations from contaminating the primary interest recommendations, thereby maintaining recommendation accuracy while managing complexity through modular design.

The patent implements a universal recommendation engine architecture that can switch between different recommendation models based on the active mode. The system maintains a first recommendation model for primary account recommendations and a second recommendation model for private interactive mode, allowing the same underlying infrastructure to serve multiple functions with different recommendation strategies, thereby managing complexity through multi-functionality.

The patent introduces a mode management mechanism that acts as an intermediary between the user, the private interactive mode, and the recommendation models. This mediator handles the switching between recommendation models, manages the isolation of data, and controls the deletion of private mode information, thereby coordinating the complex interactions between multiple recommendation models without requiring complete system redesign.

The patent applies local quality by creating a private interactive mode with distinct data handling characteristics. Within this private mode, user interaction data is retained and used for personalization purposes, but this data isolation ensures that the information does not affect the primary account's recommendation profile. The private mode has localized data retention policies that differ from the main system, enabling personalization without compromising overall privacy.

The patent implements a discard-and-recover mechanism for private interactive mode data. User interaction data generated during private mode sessions is retained temporarily to provide personalized recommendations within that mode, but is automatically deleted when the user exits the private mode or logs out. This approach allows the system to utilize data for personalization during the active session while ensuring privacy by discarding the data afterward, preventing permanent storage and misuse.