Feature Model Instance Mining For Software Product Line Engineering

The software product line is a large-scale software reuse method that is an effective way to develop a set of applications with similar and similar requirements.In the development of software product lines,domain requirements can express the commonality and variability of the domain in the form of feature models,and by designing efficient single-objective or multiobjective optimization feature selection algorithms,the solution set of the algorithm can be greatly satisfy customer needs.There are some problems in the software product line field that need to be solved urgently.On the one hand,because of the small number of feature models that exist in reality,researchers often use random feature models or individual feature models found in other literatures.On the other hand,with the increasing demand of users,it is necessary to weigh multiple targets in the development of software products.However,the multi-objective optimization feature selection algorithm in the existing literature can find more non-dominated solutions,but it may contain Inferior solutions such as local optimal solutions or dominating solutions.Therefore,according to the above two problems,how to mine more real feature models and design a multi-objective optimization feature selection algorithm to avoid local optimal solutions and dominating solutions becomes more important.In view of the above two problems,the main results and conclusions of this paper are as follows:(1)Mining and verifying the feature models existing in reality.In this paper,the data in the Debian software package is used to map the dependencies between the software packages and the relationship between the feature models.The extracted feature models are filtered and verified.The results show that the newly mined feature models are effective.(2)Design and implement a better multi-objective optimization feature selection algorithm.In this paper,by improving the solution process of multi-objective integer programming algorithm,and using existing datasets and newly mined dataset experiments,the results show that this method not only effectively avoids the generation of local optimal solutions and dominating solutions,but also has better performance than the existing algorithm.