Classification method for airborne system software defects

Abstract

The invention provides an intelligent classification method for airborne system software defects, and comprises the following steps: providing a word segmentation dictionary oriented to the aviation field to realize accurate word segmentation of airborne software; then, proposing an improved LDA topic model integrated with airborne software demand characteristics, acquiring implicit topics oriented by demands such as function names and interface names, and improving topic acquisition accuracy of the LDA model. On the basis of the obtained topic model, complex field characteristics and failure laws of the airborne software defect data can be well met, finally, accurate classification and prediction of the airborne software defect data are achieved, and then the efficiency and quality of airborne software test design work are improved.

Description

technical field [0001] The invention relates to a method for classifying software defects of an airborne system, belonging to the technical field of software. Background technique [0002] Software defect classification technology is an effective way to improve defect identification and prediction. Most software defect data are natural language texts, which are irregular and ambiguous. It is difficult for computers to process and classify software defect data effectively. However, if we only rely on manual methods to classify historical defect data and realize the reuse of the same type of software defects, the workload required is relatively large, and subject to human subjective factors, it is easy to cause omissions, and it is difficult to guarantee the efficiency of defect classification and quality. Therefore, using artificial intelligence technologies such as natural language processing to intelligently classify and reuse software defect data is one of the effective...

AI Technical Summary

Problems solved by technology

[0003] Compared with general software defects, airborne software defects have more complex domain characteristics. Existing text segmentation techniques (such as conditional random field model) or general topic acquisition models (such as Latent Dirichlet Allocation topic model, namely latent Dirichlet Lai distributed topic model, hereinafter referred to as the LDA topic model), it is difficult to accurately segment the text or obtain the topic of the airborne software defect data

In addition, the existing data classification technology based on topic model is mostly unsupervised learning technology, which has less consideration for the prior statistical characteristics of data, which affects the accuracy and recall rate of airborne software defect data classification

At the same time, airborne users have higher requirements for quality factors such as the overall functional performance of equipment, while software testing generally focuses on the compliance of the whole machine and subsystems with requirements, as well as the implicit requirements of software task reliability and user experience. Subsystem-level black-box software testing requires a lot of manpower and resources, and on-board software testing is inefficient

Images