Cross-Media Entity Recommendations for New Content
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional co-click analysis is limited to media items of the same type, making it ineffective for recommending related items for new media content since no user interactions have been recorded, leading to a delay in providing recommendations and potential loss of business for websites hosting time-sensitive content.
Innovation Solution
A system that provides cross-media type recommendations by performing co-interaction analysis across different media types associated with common entities, using co-interaction data from various media items such as news articles, videos, purchases, and social media content to determine relatedness scores and recommend relevant items.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional co-click analysis is used to determine related media items, then the accuracy of recommendations is improved, but the time required to provide recommendations increases due to waiting for sufficient user interactions
Solution Approach 1:
The patent applies preliminary action by pre-computing and storing co-interaction statistics between entities across different media types before they are needed for recommendations. When a new media item is added, the system can immediately retrieve pre-computed entity relationships rather than waiting for user interactions to accumulate, thus providing timely recommendations while maintaining accuracy through pre-analyzed data
Solution Approach 2:
The patent transitions from analyzing co-interactions within the same media type to cross-media type entity relationships. By extracting entities from media items of different types (videos, articles, purchases) and analyzing their co-interaction patterns across media boundaries, the system creates a new dimensional approach that enables recommendations for new items without requiring item-specific interaction history
2Device complexity
If co-click analysis is limited to media items of the same type, then the complexity of the system is reduced, but the versatility of recommendations is limited
Solution Approach 1:
The patent introduces entities as intermediary elements that connect different media types. Instead of directly analyzing relationships between diverse media items, the system extracts entities from each item and uses these entities as mediators to establish cross-media relationships. This intermediary approach enables versatile cross-type recommendations while keeping the system manageable by standardizing relationships through entity annotations
Solution Approach 2:
The patent implements universality by creating a unified entity-based framework that handles multiple media types through a common approach. The same entity extraction and co-interiction analysis methodology applies to videos, articles, purchases, and other media types, allowing the system to provide versatile recommendations across different media domains using a single generalized system
Data Source
AI summary
Recommendations for a media item associated with a primary entity are based on co-interaction information gathered from other media content items of several different media types that are also associated with the primary entity. Co-interaction information can include, for example, co-click data for websites, co-watch data for videos, or co-purchase data for purchases. The co-interaction data is processed to determine a co-interaction score between primary media items and secondary media items. From the co-interaction scores, secondary entities associated with the secondary media items are determined. A relatedness score is determined for these secondary entities based on the aggregation of the co-interaction scores of the secondary media items they are associated with. The relatedness score indicates a determination of how related one entity is to another. The secondary entities are ranked according to relatedness score in order to determine secondary entities most relevant to the primary entity.