Contextual Relationship Graphs for Targeted Content Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for providing targeted content, such as targeted advertising, face challenges in accurately classifying recipient context and predicting relevant non-competitive advertisements across retailers, leading to inefficiencies and limitations in scalability and flexibility.
Innovation Solution
A system and method utilizing hierarchical predictive projections with contextual relationship graphs to select targeted content based on request-associated attributes, enabling the selection of personalized and non-competitive advertisements without human intervention, and adapting to evolving business needs through product recommendation boosting and ensemble learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If the bucket of words approach is used for context classification, then the processing speed is fast, but the accuracy of targeted content delivery deteriorates due to lack of contextual relationships
Solution Approach 1:
The patent segments the context classification process into two distinct phases: (1) a fast bucket-of-words phase that identifies candidate context terms through statistical word frequency analysis, and (2) an accuracy-enhancing phase that uses contextual relationship graphs to verify and refine these candidates by examining semantic relationships between words. This segmentation allows the system to benefit from both the speed of simple frequency counting and the accuracy of contextual analysis.
Solution Approach 2:
The patent introduces contextual relationship graphs as an intermediary structure that bridges the gap between simple word frequency counting and accurate context understanding. These graphs store pre-computed semantic relationships between words, allowing the system to quickly verify whether candidate context terms actually have meaningful relationships with other words in the document, thereby improving accuracy without requiring full natural language processing.
2Measurement precision
If natural language processing is used for context classification, then the accuracy of targeted content delivery improves, but the processing speed deteriorates due to prefiltering requirements and human involvement
Solution Approach 1:
The patent performs preliminary action by pre-computing and storing contextual relationship graphs offline before real-time advertising operations. These graphs contain pre-analyzed semantic relationships between words that would otherwise require expensive natural language processing computations. During real-time operation, the system simply queries these pre-computed graphs rather than performing full NLP analysis, achieving both high accuracy and fast processing.
Solution Approach 2:
The patent extracts only the essential contextual relationship information needed for advertising relevance from full natural language processing, storing these extracted relationships in contextual relationship graphs. This extraction approach captures the most important semantic relationships while discarding unnecessary linguistic analysis, thereby maintaining accuracy for advertising purposes while dramatically reducing processing requirements.
3Measurement precision
If classification trees for web pages and advertisements are not congruous, then matching difficulty increases, but adding human involvement to resolve mismatches decreases efficiency
Solution Approach 1:
The patent changes the parameter used for matching from rigid classification tree structures to flexible contextual relationship graphs. Instead of requiring exact matches between predefined classification categories, the system uses semantic relationship strength as a matching parameter, allowing it to find relevant advertisements even when classification trees don't align perfectly. This parameter change enables automated matching based on actual contextual relevance rather than structural congruence.
Data Source
AI summary
A system and process for incorporating recommendation boosting in an automated recommendation system includes presenting a user with a visual electronic interface adapted to receive recommendation boost instructions regarding a boost subject, receiving recommendation boost instructions via the visual electronic interface, wherein the recommendation boost instructions indicate how strongly the boost subject should be recommended or suppressed from being recommended, receiving a set of recommendations from one or more automated product recommendation systems, wherein each recommendation system utilizes one or more selection models or user models and modifying the set of recommendations according to the recommendation boost instructions.