Sparse User Media Recommendation via Entity Vector Dot Products
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current content recommendation systems fail to provide robust and personalized recommendations for users with sparse viewing histories, as they rely on Content Based or Collaborative Filtering methods that struggle to generate diverse and accurate recommendations.
Innovation Solution
A method that creates a media entity matrix and user preference vector, using attributes like keywords, genres, and themes, and applies dot products to determine entity rankings, weighting entries by a user's epoch to prioritize relevant content, thereby providing personalized recommendations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If Content Based or Collaborative Filtering methods are used for recommendation, then recommendations can be generated for users with sufficient consumption history, but recommendation accuracy and personalization deteriorate for users with sparse viewing histories
Solution Approach 1:
The patent introduces an entity-based intermediary approach where recommendations are generated through entity rankings rather than direct user-content matching. By creating entity vectors that represent production entities (directors, actors, producers) and computing dot products between user preference vectors and entity vectors, the system mediates the recommendation process to provide accurate suggestions even for users with limited consumption history
Solution Approach 2:
The patent transforms the recommendation problem by changing parameters from direct content attribute matching to entity-based scoring. It introduces epoch factors that weight scores based on temporal relevance, and transforms content attributes into entity rankings through dot product computations, enabling effective personalization for sparse users
2Adaptability or versatility
If traditional recommendation methods are applied, then system complexity remains manageable, but the ability to transform attributes into relevant affinity metrics is insufficient
Solution Approach 1:
The patent replaces traditional mechanical recommendation systems with a mathematical transformation approach. It substitutes direct content matching mechanisms with dot product computations between vectors, transforming content attributes into entity rankings through linear algebra operations. This substitution enables sophisticated attribute transformation while maintaining computational efficiency
Data Source
AI summary
A processor may create a preference vector by generating preference vector entries in fields that correspond to content attributes associated with at least one media program presented via a media device, each entry including a playback duration and a year. The processor may create a media entity matrix of entity vectors for entities associated with production of media programs. For each entity vector, the processor may generate matrix entries in fields that correspond to content attributes associated with at least one media program associated with the entity, each matrix entry including a year and a score, the score weighted by an epoch factor based upon a time span associated with the at least one media program presented via the media device. The processor may determine entity rankings based upon dot products between the preference vector and the entity vectors and provide a recommendation based upon the rankings.