مسعود اختيارزاده

The names used in the software source code, especially the classes and methods, play a significant role in its readability. Class and method names should reflect the concept and purpose of the classes and methods in the source code. Method naming is one of the most important factors in program understanding, which directly affects quality features such as readability, testability, and comprehensibility. Identifiers set by developers make up 70% of an application. Among the different identifiers, method names need more attention because methods are the smallest meaningful and reusable pieces of code. Therefore, helping the developers to choose the appropriate method names automatically reduces the development cost and time. This thesis proposes the use of software criteria as features to show the performance and structure of methods and classes to be compared. We use source code metrics to extract methods and classes in a large collection of high-quality projects and also calculate textual similarity using the TF-IDF algorithm. Vectors are mostly used as criteria for suggesting suitable names for a given method or class. For this purpose, the K-Nearest Neighbor algorithm is used to determine the k-most similar methods or classes to the desired methods or classes in an unsupervised manner. We present a formula to incorporate the similarities of the methods in the respective classes as a factor influencing the final similarity score. Experiments on 800 Java projects with nearly 4,000,000 methods confirm that our proposed approach is more effective and efficient than the state-of-the-art Code2Vec and Code2seq approaches. In manual eva‎luations, based on experts' opinions, the Precision and Recall of our proposed methods shows an improvement of about 8.33% and 8.18%, respectively. In our automated eva‎luations, Precision and Recall improved by 4.25% and 12.08%, respectively, and in the class proposal discussion by testing on 76,037 classes, the F-score improved by 2.93%. Finally, the total runtime of the proposed model for 420,852 methods in our test set is reduced by 13,186 minutes.

نامگذاري متد , نامگذاري كلاس , پيشنهاد , شباهت كد منبع , سنجه هاي كد منبع

Method naming , class naming , Recommendation , code similarity , source code metrics

Masood Ekhtiyarzade

Dr. Saeed Parsa