Graph Neural Network Defect Location via Coverage Information
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing defect location technologies based on coverage information suffer from loss of detail in coverage information during processing, leading to low accuracy in defect location.
Innovation Solution
A defect location method and device that characterize program structure information and test case coverage information in a graph, generating a node attribute sequence and adjacency matrix, which are then input to a trained graph neural network model to output a suspicious degree list of the target program.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If traditional ranking learning model compresses the graph before learning, then the processing complexity is reduced, but the coverage information is lost which results in low accuracy of defect location
Solution Approach 1:
The patent introduces a graph neural network as an intermediary that can process graph-structured data without requiring compression. The GNN model takes the original graph structure with full coverage information as input and processes it through multiple layers of neural network transformations, thereby avoiding information loss while maintaining manageable computational complexity through the structured processing approach.
Solution Approach 2:
The patent changes the processing parameters by using the original graph structure directly as input to the neural network model rather than compressing it first. The model learns to process the full-dimensional graph data through weight matrices and activation functions, transforming the parameters of processing from compressed representation to full representation with appropriate neural network operations.
2Quantity of substance
If coverage information is compressed for processing, then the data size is reduced, but the accuracy of defect location deteriorates due to information loss
Solution Approach 1:
The patent performs preliminary action by directly feeding the complete graph structure with full coverage information into the neural network model before any processing occurs. The model is pre-trained to handle the full-dimensional input data, preserving all coverage information throughout the processing pipeline rather than compressing it beforehand.
Solution Approach 2:
The patent replaces the mechanical compression approach with a neural network-based processing system. Instead of using traditional data compression techniques that reduce data size, the solution uses a graph neural network that can efficiently process the full-dimensional graph data through learned transformations, substituting mechanical compression with intelligent neural processing.
Data Source
AI summary
The present disclosure discloses a defect location method and device based on coverage information, the method including: characterizing program structure information and test case coverage information of a target program in a graph to obtain a graph-characterized structure; generating a node attribute sequence and an adjacency matrix based on the graph-characterized structure; and inputting the node attribute sequence and the adjacency matrix to a trained graph neural network model, so that the graph neural network model outputs a suspicious degree list of the target program based on the node attribute sequence and the adjacency matrix. Through the graph-based unified information characterization, the coverage information can be saved without loss and compression, and the structure information may also be considered, thereby improving the accuracy of defect location.