Fusion-Vector Content Recommendation for Relevance and Diversity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing recommendation systems suffer from high similarity in recommended content and poor diversity, leading to user aesthetic fatigue and diminished browsing experience.
Innovation Solution
A method involving a computer device that generates fusion vectors from an input structure graph, using an interactive prediction model to determine interactive prediction values and diversity indices, selecting content nodes that balance correlation and diversity for improved recommendations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If matrix decomposition is used to obtain word embedding vectors for user and video, then interaction possibility can be predicted, but the similarity between recommended videos becomes high and diversity is poor
Solution Approach 1:
The patent introduces a multi-dimensional evaluation framework that extends beyond traditional single-dimensional similarity scoring. By incorporating user profile dimensions, video content dimensions, and interaction history dimensions into a comprehensive evaluation model, the system achieves both accurate interaction prediction and diverse recommendations. The fusion of multiple dimensional features allows the system to balance precision and diversity effectively.
Solution Approach 2:
The patent dynamically adjusts recommendation parameters based on user behavior patterns and contextual information. By changing the weighting of different features and parameters according to real-time user state and historical data, the system can maintain high prediction accuracy while introducing diversity. The adaptive parameter adjustment enables the system to switch between precision-oriented and diversity-oriented recommendation modes as needed.
2Reliability
If content recommendation focuses on high similarity items, then user needs are met, but user aesthetic fatigue increases
Solution Approach 1:
The patent implements a feedback mechanism that continuously monitors user interaction with recommended content. By analyzing user feedback signals (such as viewing duration, engagement level, and interaction patterns), the system dynamically adjusts the recommendation strategy. This feedback loop enables the system to maintain relevance through precise prediction while preventing aesthetic fatigue by introducing varied content based on real-time user response patterns.
Solution Approach 2:
The patent employs dynamic recommendation strategies that adapt to changing user preferences and contexts. Rather than static similarity-based recommendations, the system dynamically adjusts content selection based on real-time user state, temporal patterns, and contextual information. This dynamic approach ensures continued relevance while introducing variety to prevent user fatigue.
Data Source
AI summary
A recommended content selection method is performed by a computer device. The method includes: obtaining an input structure graph for representing associations between user nodes and content nodes; generating fusion vectors corresponding to nodes in the input structure graph using an interactive prediction model, a fusion vector corresponding to a target node integrates feature information of the target node and feature information of another node associated with the target node; determining, for a target user node, interactive prediction values between the target user node and the plurality of content nodes based on a fusion vector corresponding to the target user node and fusion vectors corresponding to the plurality of content nodes; and selecting a recommended content node from the plurality of content nodes according to the interactive prediction values between the target user node and the plurality of content nodes and a diversity index of a recommended content node set.