Research On Evaluation Technology Of Recommendation Results Based On API Usage

With the development of development languages and various software communities,the number of APIs has increased dramatically.In order to reduce the difficulty of using API and improve the efficiency of using API,researchers put forward a lot of API recommendation methods,which can help programmers to use API more efficiently.With the increasing of recommendation technology,it is very important for programmers to choose the right recommendation method.The evaluation of API recommendation results can provide the basis for programmers to choose recommendation technology.However,there are few evaluation methods,and mainly focus on the correctness of recommendation results,lack of attention to the quality of recommendation results,which can hardly meet the needs of developers for evaluation technology.In order to solve the above problems,this thesis proposes a recommendation result evaluation technique based on API usage.The technique is based on the correctness of recommendation results,using API usage frequency and API document information to evaluate the recommendation results.Through the empirical study of API recommendation evaluation technology,this thesis finds that the result of the correctness evaluation can hardly reflect the help of the API for the problem.This thesis also finds that the frequency of API usage can be used to calculate the usability of API,and the usability in the characteristics of API usage can be used to evaluate the recommended results.In addition,the thesis also found that the API with high frequency,which greatly affects the correct score in the recommendation results,can provide low availability.And there are About 20% of the APIs constitute about 90% of the correct recommendation results.Therefore,this thesis proposes a recommendation result evaluation technology based on API usage.This technique analyzes the impact of usability in the evaluation of recommendation results,and uses the characteristics of correctness evaluation and API usage to evaluate the recommendation results.This technique can distinguish the help of the correct API in the recommendation result to solve the problem,and evaluate the recommendation result more finely.In this thesis,two API recommendation tools based on functional description and one API recommendation tool based on code context are selected for empirical research and verification experiments,and 12186 groups of recommendation results data are collected.The experimental data show that the recommendation result evaluation technique based on API usage can quantitatively evaluate the contribution of correct recommendation results to the completion of functional requirements according to the characteristics used by API,and the evaluation results can make up for the shortcomings of correctness evaluation and are more effective.Through the efficiency comparison,it can be seen that the recommendation results evaluation technique based on API usage does not cost much more time than the correctness evaluation.