Graph Neural Network Search Accuracy via Media Graph Sampling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current search technologies face inaccuracies due to limited information in queries or media data, leading to inaccurate semantic features and search results.
Innovation Solution
A search data processing method that constructs a media search graph, samples it based on meta-paths, and uses a graph neural network to extract semantic features from query and media nodes, training the network on positive and negative node pairs to improve search accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If feature extraction is directly performed on query or media data, then the process is simple and fast, but the semantic feature accuracy is low due to limited information
Solution Approach 1:
The patent transitions from direct feature extraction in the original data space to feature extraction in an expanded dimensional space by constructing a media search graph. This graph includes query nodes, media nodes, and association nodes that capture relationships between queries and media data across multiple dimensions (user interactions, media attributes, query characteristics). The graph neural network then operates on this multi-dimensional structure to extract more comprehensive semantic features, resolving the contradiction between simplicity and accuracy.
Solution Approach 2:
The patent applies preliminary action by pre-constructing a media search graph that enriches raw query and media data with association information before feature extraction. The graph incorporates historical search data, user interactions, and media attributes in advance, so that when the graph neural network performs feature extraction, it already has access to comprehensive contextual information. This preliminary enrichment enables accurate semantic feature extraction without requiring complex post-processing.
2Measurement precision
If more information is incorporated into the search data, then the semantic feature accuracy improves, but the data processing complexity increases
Solution Approach 1:
The patent segments the complex search data processing task into distinct components represented as separate node types in the media search graph: query nodes for search queries, media nodes for media data, and association nodes for relationship information. Each node type processes and stores specific kinds of information independently. The graph neural network then aggregates information from these segmented nodes through message passing, enabling accurate feature extraction while maintaining manageable processing complexity through modular organization.
3Measurement precision
If a graph neural network is used to extract features from the media search graph, then the semantic feature accuracy improves, but the computational resources required increase
Solution Approach 1:
The patent extracts only the essential and relevant features from the media search graph using the graph neural network, rather than processing all possible information. The network extracts semantic features that are most important for search accuracy by learning to weight and select relevant attributes from the graph structure. This selective extraction reduces computational resource consumption while maintaining high semantic feature accuracy by focusing on the most informative aspects of the search data.
Data Source
AI summary
A search data processing method includes: obtaining a media search graph including a query node, a media node, and an association node; obtaining first training sample pairs, including a positive node pair and a negative node pair, from the media search graph, sampling the media search graph based on meta-paths respectively corresponding to the query and the media data, to obtain sampling sub-graphs; inputting the sampling sub-graphs to an initial graph neural network to obtain respective initial semantic features and form a semantic feature pair corresponding to the first training sample pair; and training the initial graph neural network based on a difference between semantic feature pairs of the positive node pair and the negative node pair, to obtain a target graph neural network that is configured to determine a target semantic feature corresponding to a query node or a media node.