A software quality self-adaptive measurement method based on information geometry and reinforcement learning
By combining information geometry and reinforcement learning, low-dimensional manifold coordinates of software class graphs are extracted, composite fuzzy matter-element is constructed and adaptively optimized, overcoming the limitations of traditional software quality measurement methods and achieving efficient and accurate evaluation of complex software systems.
Patent Information
- Application Number
- CN202511864814.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-11
- Publication Date
- 2026-07-03
- Estimated Expiration
- 2045-12-11
AI Technical Summary
Existing software quality measurement methods cannot adapt to the dynamic and uncertain nature of software quality evaluation. They lack geometric modeling of high-dimensional manifold spaces, traditional methods cannot accurately characterize the intrinsic structure of quality features, and they fail to fully utilize machine learning, especially reinforcement learning techniques, to achieve adaptive optimization of the evaluation process.
A method combining information geometry and reinforcement learning is adopted. Low-dimensional manifold coordinates are extracted through information geometric mapping and sparse representation to construct composite fuzzy matter-element, and the initial weights are calculated using the entropy weight method. Multiple rounds of iterative optimization are carried out through the adaptive feedback mechanism of reinforcement learning, and finally the MARCOS method is used to calculate the comprehensive evaluation value.
It enables dynamic and adaptive evaluation of software quality, improves the accuracy and discriminative power of measurement results, overcomes the limitations of traditional methods, and adapts to the nonlinear and high-dimensional quality data of complex software systems.