Temporal Graph Neural Network for Test Case Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for system testing lack the ability to predict test case attributes without access to source code and fail to account for hardware interactions, relying heavily on intuition and lacking guidance for test managers, and are unable to generalize to unseen test cases.
Innovation Solution
A computer-implemented method using a trained temporal graph neural network to analyze a corpus-level graph representing test cases, incorporating word and test step nodes with directed and undirected edges, to predict test case attributes affecting testing reports, including failure rates and interactions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning techniques like classification, regression, clustering and reinforcement learning are used for test case prioritization, then test case prediction capability is improved, but access to source code is required which is not available in system testing
Solution Approach 1:
The patent introduces an intermediary representation layer (graph structure with test case nodes and relationships) that bridges the gap between source code-based ML techniques and system testing scenarios. This intermediary allows ML models to process test case data without requiring access to the underlying source code, enabling application of advanced prediction techniques in system testing environments.
Solution Approach 2:
The patent replaces the mechanical requirement of source code access with a data-driven approach using graph neural networks. Instead of mechanically requiring source code input, the system substitutes this with processed test case metadata and relationships, allowing the same ML techniques to function in system testing where source code is unavailable.
2Extent of automation
If evolutionary algorithms and support vector machines are used for automatic test plan generation, then automation level is improved, but the methods cannot generalize to new test cases without prior performance history
Solution Approach 1:
The patent performs preliminary actions by pre-processing test case data into graph representations and training the neural network model in advance. This preliminary processing creates a robust model that can generalize to new test cases, as the model learns patterns from training data before deployment, enabling it to handle unseen test cases effectively.
Solution Approach 2:
The patent transforms test case parameters into a graph-based representation with nodes and edges, changing the parameter structure from traditional tabular data to a relational graph format. This parameter transformation enables the model to capture complex relationships and generalize better to new test cases that share similar structural patterns.
3Reliability
If test managers manually produce or design test cases, then test case quality is improved through expert intuition, but productivity is reduced due to heavy reliance on manual effort
Solution Approach 1:
The patent enables the system to serve itself by automatically generating test cases and predictions using the trained neural network model. The system processes test case data, generates predictions, and produces test plans without requiring manual intervention from test managers, thereby maintaining quality through algorithmic consistency while dramatically improving productivity.
Solution Approach 2:
The patent implements feedback mechanisms where the neural network model learns from training data and continuously improves its predictions. The system uses feedback from test case outcomes and performance data to refine its models, enabling automated generation of high-quality test cases that match or exceed manual expert quality while increasing productivity.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
There is provided a computer-implemented method for predicting one or more attributes of test cases comprising: receiving a plurality of test cases from a test case library, each test case comprising text descriptions of one or more executable test steps of the test case; generating a corpus-level graph representative of the plurality of test cases; and feeding the corpus-level graph into a trained temporal graph neural network to predict one or more attributes of each test case that affect an testing report of software, hardware, and/or system testing using the test case. There is further provided a computing system, use of computer-implemented method, computer program, machine-readable storage medium, and data carrier signal.