Evaluation Of API Recommendation System Based On API Functional Correlation

API recommendation technology is playing an increasingly important role in modern software development by recommending API methods that are suitable for current programming scenarios to program developers.Driven by the rapidly growing development demand,API recommendation technology has been rapidly released to A station,but little attention has been paid to the evaluation of API recommendation technology.At present,the evaluation method commonly adopted by researchers is borrowed from the information retrieval field or other recommendation fields,and the evaluation of correctness is the most concerned evaluation result of researchers and users.But in this thesis,through the empirical study found that the current API recommended system exist the following problems: the correctness of the assessment of the correctness of evaluation are usually qualitative,by will recommend the results match the real results and recommend judgment result is correct,but the biodiversity characteristics of program development determines to achieve the same API may be not only a kind of functional demand,namely do not match the API to a certain extent also help complete functional requirements.In order to solve the above problems of the evaluation technology,this thesis proposes a recommendation result evaluation technology based on API functional relevancy.Based on the functional relevancy between the recommendation API and the correct API,the contribution degree of the recommendation API is quantitatively calculated,and this contribution degree is included in the correctness calculation of the recommendation result.This thesis invited 15 developers to participate in a survey on the use of API recommendation assessment techniques,and the survey results showed that the unmatched results with functional relevance to the real API can help users complete functional requirements.In this thesis,based on API functional relevancy,the existing correctness evaluation method is improved,and the quantitative functional relevancy value is used to supplement the qualitative correctness index.In order to verify the effectiveness of the evaluation technique,this thesis constructs an evaluation data set containing 6141 recommendation scenarios,and generates the recommendation results to be evaluated through different recommendation parameter configurations.Experimental results show that the recommendation result evaluation technology based on API functional relevance in this paper can identify the contribution of the mismatch API,and the evaluation of the merits and demerits of different recommendation systems is more in line with the manual evaluation.In addition,the resources expended in space and time are within the acceptable range,which helps application developers to choose a good API recommendation system.