Analyzing source code identifiers for code reuse using NLP techniques and WordNet

Abstract

Massive amount of source codes are available free and open. Reusing those open source codes in projects can reduce the project duration and cost. Even though several Code Search Engines (CSE) are available, finding the most relevant code can be challenging. In this paper we propose a framework that can be used to overcome the above said challenge. The proposed solution starts with a Software Architecture (Class Diagram) in XML format and extracts information from the XML file, and then, it fetches relevant projects using three types of crawlers from GitHub, SourceForge, and GoogleCode. Then it finds the most relevant projects among the vast amount of downloaded projects. This research considers only Java projects. All java files in every project will be represented in Abstract Syntax Tree (AST) to extract identifiers (class names, method names, and attributes name) and comments. Action words (verbs) are extracted from comments using Part of Speech technique (POS). Those identifiers and XML file information need to be analyzed for matching. If identifiers are matched, marks will be given to those identifiers, likewise marks will be added together and then if the total mark is greater than 50%, the .java file will be considered as a relevant code. Otherwise, WordNet will be used to get synonym of those identifiers and repeat the matching process using those synonyms. For connected word identifiers, camel case splitter and N-gram technique are used to separate those words. The Stanford Spellchecker is used to identify abbreviated words. The results indicate successful identification of relevant source codes.