Research On Code Recommendation System Based On The Analysis Of Code Base

In the daily programming,after developers inputed expression and dot,the intergrated development environment(IDE)prompt all possible proposals of API besed on static analysis.Howerver,the existing IDE just recommend all mothods which are accessible and sorted by the letter of the alphabet.The existing class contains hundreds of API methods that can be accessed.Therefore,combined the conext of the programming and the history data of web to reduce useless API metod in the tips and make the relevant API methods ranking near the top in completion prompting bar is a effectively good work of reduce time when developers are programming.In this paper,our study includes the processing and analysis of stack overflow offline data dump and recommendation of API method based on association rules,specific work is as follows:1.Extracted the relevant code information from Stack Overflow website and then built API method use pattern database.Analyzed the post table and extracted the code base from the content of answer.Got the API instance database after pasrsed the code base to abstract syntax trees(AST).2.Counted the popularity indicator of the code of the relevant Q&A page then established the API popularity indicator databse.Proposed and tested API methode recommender based on popularity indicator.3.Because of the defect which the API mehod recommder don't consider the context of the programming environment and improve the way of the input and the rule discovering in Apriori algorithm via the features of strongly typed and object-oriented in Java programming language.Proposed a API method recommender combined popularity indicator and association rule.Implemented the recommder in the form of Eclipse plug-in.Tested the precision and the recall between our recommder and other papers presented API method recommender.Compared with with CSCC and BCC and BMN,experimental results show that the precision of Top-1 and Top-3 of our method is better than CSCC and worse than BMN.The precision of Top-10 is best than other mothd.The recall of our method is less than BMN and BCC and more than CSCC.