Automatic Classification Of Non-Functional Requirements In User Reviews

In recent years,the leading App distribution platforms(such as Apple App Store)have achieved stupendous success.Users all over the world download and use millions of mobile applications every day.Mobile applications released on such platforms have millions of potential customers around the world.However,for mobile application developers,publishing applications to the App store is just the beginning of the application lifecycle.In fact,developers are under pressure to constant improving the quality of their applications.Developers need to continuously improve their applications to compete with similar applications.On the one hand,software testing can be used to find a bug and fix it.But,software testing not only consumes huge manpower,but also is difficult to implement millions of users directly using applications on different environments and heterogeneous mobile devices.On the other hand,developers can also obtain feedback from mobile application users by conducting surveys,and get advices or suggestions on how to optimize the application to meet user needs.In addition to the above two methods,valuable user feedback exists in the mobile application store:user reviews.However,manually reading user reviews and verifying it contains useful information(such as suggestions for new features)or not consumes considerable human resource for mobile applications that receive lots of reviews per day.From the perspective of software engineering,the acquisition and analysis of requirements play a crucial role in the success or failure of software.In general,requirements are composed of functional requirements and non-functional requirements.Non-functional requirements are crucial for user satisfaction of software quality and success or failure of software.On the other hand,improper handling of non-functional requirements can increase the human,time,and costs of software development.In this thesis,user reviews were automatically classified into six categories:reliability,usability,portability,performance efficiency,functional requirements,and others from the perspective of non-functional requirements classification,and AUR-BoW(Augmented User Reviews)which uses contextual text semantics to enhance user reviews is proposed to improve non-functional requirements classification performance of user reviews.AUR-BoW is composed of three main steps,first,Word2Vec is used to calculate the similarity between words.Then the similarity between word and user review is calculated,and finally words with higher similarity are selected to expand the user review.In order to evaluate the effectiveness of the proposed method,we select user reviews from iBooks in Apple App Store and WhatsApp in Google Play Store as experimental dataset.We combined four classification techniques BoW,TF-IDF,chi2,AUR-BoW with three machine learning algorithms Naive Bayes,J48,Bagging.We conducted experiments to compare the F-measure of the classification results through all the combinations of the techniques and algorithms.During the experiment process,user reviews are first pre-processed,and then the user reviews are featured by classification techniques.After that,a machine learning algorithm is used to build a classification model.Finally the user reviews are finally divided into one of six categories.The experiment shows that AUR-BoW has the best classification performance among the four classification techniques.When combined with Bagging,the F-measure is highest(0.718).The experiment results indicate that using contextual semantic information to enhance user reviews can improve the classification performance of non-functional requirements in user reviews.