Cluster-Augmented Graph Convolutional Network for Media Retrieval
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional personalized content retrieval systems fail to capture complex non-linear user preferences and item similarities due to limitations in feature extraction methods, leading to sub-optimal performance in recommending electronic media.
Innovation Solution
The Cluster-Augmented Graph Convolutional Network (CA-GCN) framework enhances user and item embeddings by leveraging user-item interaction data and item-item similarities through clustering, propagating information between users and items to generate representative embeddings that reflect high-level semantics and user preferences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional feature extraction methods are used to extract user preferences, then the system can process user behavior data, but it fails to learn complex non-linear relationships from item features and user similarities
Solution Approach 1:
The patent replaces traditional mechanical feature extraction methods with a neural network-based Graph Convolutional Network (GCN) system. The GCN automatically learns complex non-linear relationships from user-item interaction graphs, substituting manual feature engineering with automated deep learning approaches that capture intricate patterns in user behavior and item similarities.
Solution Approach 2:
The patent combines multiple data sources and representation methods into a unified embedding space. It integrates user behavior data, item features, and graph structure information into composite user and item embeddings, creating a richer representation that captures both explicit and implicit relationships between users and items.
2Loss of information
If methods focus mostly on extracting user embeddings, then user preferences can be represented, but item features are not enhanced
Solution Approach 1:
The patent merges user embedding extraction and item feature enhancement into a unified GCN framework. The same graph convolutional operations that learn user preferences also refine item embeddings by propagating information across the user-item interaction graph, ensuring both user and item representations are simultaneously optimized without separate processing pipelines.
Solution Approach 2:
The GCN model serves multiple functions: it extracts user embeddings, enhances item features, and captures user-item interaction patterns all within a single unified framework. This multi-functional approach eliminates the need for separate processing systems for user and item representations.
3Reliability
If user-item interaction data is sparse, then the graph model can be simplified, but the retrieval performance deteriorates
Solution Approach 1:
The patent applies preliminary clustering to items based on their features before constructing the user-item interaction graph. This pre-processing step groups similar items together, allowing the GCN to leverage cluster-level statistics and propagate information more effectively even when individual user-item interactions are sparse, thereby improving retrieval performance with limited data.
4Adaptability or versatility
If traditional content retrieval systems are used, then the system structure is simple, but personalized recommendations based on user behavior are limited
Solution Approach 1:
The patent replaces traditional rule-based or keyword-matching retrieval systems with a neural network-based GCN approach. This substitution enables the system to automatically learn complex user preferences and item similarities from interaction data, providing sophisticated personalization capabilities that far exceed traditional methods despite the increased model complexity.
Data Source
AI summary
The present disclosure relates to a retrieval method including: generating a graph representing a set of users, items, and queries; generating clusters from the media items; generating embeddings for each cluster from embeddings of the items within the corresponding cluster; generating augmented query embeddings for each cluster from the embedding of the corresponding cluster and query embeddings of the queries; inputting the cluster embeddings and the augmented query embeddings to a layer of a graph convolutional network (GCN) to determine user embeddings of the users; inputting the embedding of the given user and a query embedding of the given query to a layer of the GCN to determine a user-specific query embedding; generating a score for each of the items based on the item embeddings and the user-specific query embedding; and presenting the items having the score exceeding a threshold.