Code Similarity Prediction Model for Industrial Management Features Based on Graph Neural Networks.

Area of Science:

• Computer Science
• Software Engineering
• Artificial Intelligence

Background:

• Industrial management software code presents challenges for similarity computation due to limited system API calls and extensive custom variables.
• Traditional text-based and neural network methods struggle with the unique characteristics of this type of code.

Purpose of the Study:

• To propose a novel model, FSPS-GNN, for effective code similarity computation in industrial management software.
• To address the limitations of existing methods in analyzing complex code structures.

Main Methods:

• Developed the FSPS-GNN model utilizing graph neural networks (GNNs).
• Categorized code features into outer and inner graphs.
• Employed a four-stage process: feature embedding, enhancement, fusion, and similarity prediction.
• Utilized differently structured GNNs in embedding and enhancement stages to improve feature interaction.

Main Results:

• Achieved an average precision of 87.57% and an F0.5 Score of 89.12% on code from three open-source projects.
• Demonstrated a lower Mean Squared Error (MSE) compared to existing GNN-based models (0.0041 to 0.0266 lower).
• Showcased a higher F0.5 Score (3.3259% to 6.4392% higher) than comparative models.

Conclusions:

• The FSPS-GNN model effectively computes code similarity for industrial management software.
• The study expands the applicability of GNNs in software engineering and code analysis.
• Findings provide valuable insights into addressing code-similarity challenges.