Skill-Enriched Graph Neural Network Training for Cold-Start Recommendations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional recommender systems using graph neural networks (GNNs) generate inaccurate recommendations for new and inactive users due to sparse graph input data, leading to the 'cold start' problem, which inhibits the ability of new users to engage effectively with online systems.
Innovation Solution
Incorporating user skills as nodes in the input graph enhances the GNN training process, allowing for more comprehensive and connected embeddings, thereby improving the accuracy and stability of recommender systems for new and inactive users.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional GNNs are used for recommender systems, then the system structure is simple, but the recommendation accuracy deteriorates for new and inactive users due to sparse graph input data
Solution Approach 1:
The patent applies preliminary action by pre-processing user data to extract skill information before constructing the graph. Skills are identified and incorporated into the graph structure in advance, allowing the GNN to work with enriched data even when user interaction history is sparse. This preliminary enrichment of the graph with skill nodes and edges enables better recommendation accuracy for new users without requiring complex real-time processing.
2Measurement precision
If more graph data is collected to improve recommendations, then recommendation accuracy improves, but the time to collect and process data increases
Solution Approach 1:
The patent extracts skill information from user profiles and interaction data, separating this valuable semantic information from the raw interaction logs. By extracting and incorporating skills as explicit nodes in the graph, the system leverages existing data more effectively without requiring additional data collection. This extraction approach improves recommendation accuracy while avoiding the time cost of collecting more raw interaction data.
3Manufacturing precision
If user skills are incorporated as nodes in the graph, then node embedding quality improves, but the graph complexity and computational requirements increase
Solution Approach 1:
The patent segments the graph into distinct node types including user nodes, skill nodes, and item nodes, with different properties and relationships for each type. This segmentation allows the GNN to process different types of information through specialized embedding mechanisms. Skill nodes serve as intermediary representations that capture user capabilities, enabling higher quality embeddings without requiring a monolithic complex graph structure. The segmented approach manages computational complexity by treating different node types with appropriate methods.
Data Source
AI summary
Methods, systems, and apparatuses include training a graph neural network. Content items are received for a user of an online system, the content items including skills. An input graph is generated using the content items, the input graph including nodes and edges linking the nodes. The input graph is sampled using a source node and skills to generate a first computational graph. The input graph is sampled using a target node and skills to generate a second computational graph. A source node embedding is generated by encoding the first computational graph. A target node embedding is generated by encoding the second computational graph. A prediction score is calculated by decoding the source node embedding and the target node embedding. Weights of the graph neural network are updated using the prediction score.