Temporally Sequenced Content Recommendations from Affinity Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computer-implemented recommender systems and search engines operate independently, failing to deliver the most useful information to users, necessitating a system that integrates behavioral-based indexing and contents-based indexing to provide personalized recommendations.
Innovation Solution
A system and method that combines behavioral-based indexing and contents-based indexing to infer user interest and expertise levels, mapping topical areas to object informational elements, and selecting objects for delivery based on relevance values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If behavioral-based indexing and contents-based indexing operate independently, then each system can maintain its own simple structure, but the overall system cannot provide personalized and context-aware recommendations
Solution Approach 1:
The patent merges behavioral-based indexing and contents-based indexing into a unified recommendation system. The system integrates user behavioral data (clicks, views, interactions) with content analysis (text, metadata, contextual information) to generate personalized recommendations. This combination allows the system to leverage both user-specific patterns and content-relevance metrics, achieving personalization while maintaining manageable complexity through modular architecture.
Solution Approach 2:
The recommendation system is designed to perform multiple functions: it conducts behavioral analysis, performs content indexing, evaluates user interests, and generates personalized recommendations. By creating a multi-functional system that handles both behavioral and content-based tasks within a single framework, the patent achieves versatility without requiring completely separate independent systems, thus balancing adaptability with operational simplicity.
2Measurement precision
If the system integrates behavioral-based indexing and contents-based indexing, then personalized recommendations can be provided, but the system complexity increases
Solution Approach 1:
The patent segments the recommendation system into distinct modules: a behavioral indexing component that processes user interactions, a contents-based indexing component that analyzes content properties, and a recommendation generation component that synthesizes both. This segmentation allows each module to specialize in specific tasks, improving measurement precision through focused analysis while managing overall system complexity through clear module boundaries and independent operation.
Solution Approach 2:
The system introduces an intermediary recommendation engine that bridges behavioral-based indexing and contents-based indexing. This intermediary component receives inputs from both indexing systems, evaluates user interests against content relevance, and generates final recommendations. The intermediary acts as a mediator that integrates the two indexing approaches without requiring direct complex interactions between them, thus maintaining recommendation accuracy while controlling system complexity.
3Adaptability or versatility
If user interest levels are inferred from behavioral information, then personalized recommendations improve, but more data processing is required
Solution Approach 1:
The system performs preliminary action by pre-processing and indexing user behavioral data as it is collected, organizing it into structured formats that facilitate rapid interest inference. Behavioral events (clicks, views, interactions) are immediately categorized, timestamped, and stored in an optimized structure. This preliminary organization enables efficient real-time or near-real-time inference of user interests without requiring intensive processing during recommendation generation, thus improving adaptability while maintaining productivity.
Data Source
AI summary
A temporally sequenced content recommender method and system generates a first vector of affinities from usage information comprising durations of users' attention toward temporally sequenced content. A second vector of affinities is generated by applying a computer-implemented neural network to content objects. The similarity of the vectors of affinities is determined, which informs the generation of recommendations that are delivered to a user. Beneficial serendipity may be incorporated within the generation of the recommendations by, for example, evaluating contrasting affinities in user affinity vectors, by evaluating levels of available behavioral information for users, or by applying randomized or probabilistic methods. Explanations for the recommendations may be provided to users that include inferred user preferences.