API Recommendation For Software Questions In Crowd Knowledge Platform

Crowd knowledge platform is a hot research topic in software engineering area in recent years.In a crowd knowledge platform,developers discuss software questions by asking and answering questions.These questions contain a lot of knowledge for developers to develop software and solve a software question.However,the questions in a crowd knowledge platform usually requires human to answer,which may make a delay for developers to achieve the correct answers.It turns to be a question on how to answer the questions in a crowd knowledge platform automatically.This thesis researches the problem on Application Program Interface(API)recommendation for software questions in crowd knowledge platform,in order to help developers automatically answer questions they submitted in a crowd knowledge platform.After analyzing the questions in crowd knowledge platform,this thesis proposes a fine-grain recommendation algorithm.This algorithm recommends APIs by analyzing the similarity between the question and APIs,as well as the code-like terms in a question.This algorithm identifies the code-like terms in a question and link these terms with their potential related APIs.After that,this algorithm also tries to rank these potential related APIs according to their relatedness with the question.This algorithm analyzes the characters of a question from multiple perspectives and narrows down the correct APIs according to the code-like term analysis,which improves the precision on recommending APIs for a question.This thesis analyzes the performance of the algorithm by a classical crowd knowledge platform(Stack Overflow).Experiments show that the algorithm in this thesis can correctly recommend APIs for developer questions and outperforms comparison algorithms in different evaluation metrics.Meanwhile,the experiment also analyzes the performance of the code-like term analysis method proposed in this thesis.This method also helps the comparison algorithm improve their performance.At last,this thesis discusses the threat to validity of the research and analyzes the role of each component of the proposed algorithm.The thesis also gives some suggestions on using the proposed algorithm.