Mobile App Recommendation System Category Relevance

The patent segments the recommendation approach into two independent components: (1) relevance-based recommendations using user history logs for accurate matching, and (2) category-based recommendations using pre-defined application categories to ensure diversity. By dividing the recommendation system into these separate segments that work together, the patent achieves both accuracy and diversity without one compromising the other.

The patent performs preliminary action by pre-defining application categories and establishing category relationships before the recommendation process begins. This preliminary structure allows new applications to be immediately categorized and recommended based on their category associations, even before any user history data is available. The pre-established category framework ensures new applications are not excluded from recommendations due to lack of usage history.

The patent applies local quality by differentiating the treatment of different application categories based on their relevance scores. Highly relevant categories receive priority weighting in the recommendation algorithm, while less relevant but still related categories are included with lower weights. This localized differentiation ensures that the most relevant categories drive the recommendations while maintaining broader coverage through inclusion of peripheral categories.