Research On API Recommendation For Developing Android APPs

With the rapid development of software technology,the requirements of software functionalities have also been changing fast.More and more software functionality requirements of various kinds are being brought up by developers,and the complexity for implementing certain functionalities is growing too.In order to improve software development efficiency of developers,an increasing number of software libraries have been developed to help developers to implement similar functionalities.Software development become more and more dependent on existing libraries and developers often utilize various kinds of search engines to look for suitable library APIs to use.There have been many studies on recommending relevant libraries and APIs for developers when they are writing code according to certain functional requirements.However,few studies have paid attention to an important need particularly associated with eventdriven programming,such as the Android framework.Based on our empirical study,many Android developers not only need to know which functional APIs may be invoked to implement the functionality,but only need to know where to place their implementation code according to events that may be triggered within the framework.Although a lot of efforts have been made to recommend functional APIs to developers,few studies haven't even considered the need for recommending event callbacks for development under event-driven frameworks like Android.To further illustrate and address this common need for Android developers,this paper develops an API recommendation engine for Android application development that can recommend both functional APIs for implementing the functionality and the event callbacks that are suitable to be overridden to contain the functional code.Our approach is to build domainspecific API description databases indicating the correlation among various functionalities and functional APIs and event callbacks.The techniques are based on domain-specific parsing of code snippets and natural language processing of texts contained in Android development tutorials and SDK documents.Then,textual and code similarity measurements are adapted for recommending relevant APIs.We have evaluated our prototype with more than 1,500 questions related to Android development from StackOverflow.com.We randomly selected 370 questions as our test data,and evaluated our prototype using different options for settings like database selection and query input to get the best settings for the large scale evaluation on the 1500 plus questions.According to our evaluation we demonstrated that recommending event callbacks for event-driven framework like Android is applicable,and our prototype has promising results in recommending both functional APIs and event callbacks.