The Research Of Software Maintainability Evaluation Via Topic Modeling

Because the software maintainability as the most important software quality characteristic measures the difficulty degree of the software maintenance which is directly related to the cost of developing activities, thus ensuring the software maintainability is the key of cost controlling. Driven by the fact that evaluates the software quality as an important way to assure the software quality a number of practical quality evaluation models have been proposed. However there are still some shortcomings in these models by analyzing the procedure of modeling detail, such as: determining mapping subjectivity. As software source code with some characteristics of natural language and there are some similarities between the computing framework of software maintainability evaluation model and the framework of topic mode, so we innately propose that using discrimination probabilistic latent semantic analysis model(DPLSA) to calculate the mapping among the metric unit, the quality sub-characteristics and the maintainability of source code files. Our method overcomes the shortcoming which measure the complex mapping between the metric and the maintainability objectively. The main work and innovation of this article include:(1) Statistical analyzing and discerning the metrics data which come from fourteen Java open source projects which applied to different areas, the total version is 400 and the source files over 410,000, then constructed our benchmark.(2) Proposing to use DPLSA model to calculate the mapping innately. DPLSA model can be used to calculate the complex mapping among the metrics, the quality sub-characteristics and the maintainability of source code files. And then implement the coding of core algorithm in MATLAB.(3) The complex mapping which calculate by DPLSA model as input and using the bottom-up aggregation method to build the software maintainability evaluation model via topic modeling.(4) Using our software maintainability evaluation model to calculate the maintainability of ten open source projects which include multiple versions of ten open source projects maintainability. Then using Spearman correlation analysis method to analyze the correlation relationship between code change and the maintainability which calculate by our model. The correlation analysis results demonstrated the effectiveness of our model. Thus indirectly proved that the correctness of the complexity mapping which calculated by DPLSA model.Visualizing the maintainability calculation results, we can see that the most projects have the trend that the maintainability of projects becomes better as projects progresses. The validation results showed that the Badness value of maintainability and quality sub-characteristics are positive correlate with code change of project, and the average correlation coefficient value on ten assessment project is 0.5137 which demonstrated the effectiveness of our model. This paper as an innovative research will provide a new perspective for other software quality evaluation research.