Research On A Combined Method For Usage Of NLP Libraries For API Recommendation

Application Programming Interface(API)plays an important role in software development.During the process of software development and maintenance,developers can significantly improve the development efficiency with the help of APIs.As the number of APIs has increased dramatically,it has become more difficult for developers to use them.API recommendation techniques help developers quickly find the APIs they need.Researchers need to analyze all kinds of software documents in API recommendation,from which they can extract information such as functional descriptions and usage examples of APIs.Information extraction usually relies on Natural Language Processing(NLP)technology.Therefore,improving the analytical capability of NLP libraries can optimize the application of API recommendation techniques to software documents,which is of great significance to the research of API recommendation technology.In order to analyze software documents more accurately,this thesis proposes a combined method for usage of NLP libraries for API recommendation.The main work of this thesis is as follows:(1)A combined method for usage of NLP libraries is studied in this thesis,which uses the complementarities of different NLP libraries to obtain accurate results.The problem of selecting NLP libraries in the process of recommendation is solved.The combined method mainly consists of two steps,i.e.,document-level selection of NLP library and sentence-level overwriting.Firstly,the overlapping degree of natural language processing results is used as a metric,and the NLP library with the highest overlapping degree is selected as the document-level NLP library.Then,the processing results of the document-level NLP library are analyzed sentence by sentence.The sentence-level processing results with low scores are overwritten by optimal results of other NLP libraries to achieve sentence-level overwriting.This thesis combines the two steps above to realize the complementary advantages of different NLP libraries.(2)This thesis applies the combined method for usage of NLP libraries to API recommendation.Com NT,a tool based on the combined method for usage of NLP libraries for API recommendation,is designed and implemented.This tool is used for tokenization and part-of-speech tagging for software documents in API recommendation.More accurate text processing results can be obtained.The tagging of API features can be improved.As the result,the accuracy of API recommendation can be improved.(3)4 widely used NLP libraries and 200 software documents are selected to evaluate the combined method and its application in API recommendation scenarios.The results show that the combined method can outperform a single NLP library in terms of accuracy.And the recommendation results of the API recommendation technology based on the combined method for usage of NLP libraries can get higher accuracy than a traditional API recommendation technology.