A Novel Recommendation Approach for API Sequences

API is a common approach to utilizing third-party libraries in software development. Nonetheless, to correctly call APIs, developers need to carefully learn the API documentation and search for code samples to understand their usages. In order to efficiently and effectively support developers in properly calling APIs, this paper introduces MFLUTE, a new approach to exploit program analysis and statistical language model to recommend APIs to call. Specifically, for a given partially completed code (context), MFLUTE employs program analysis techniques to analyze the context and generate a list of potential API candidates. Then, we use a language model to rank and suggest the most promising ones. Our experimental results on a large dataset of real-world projects show that MFLUTE can correctly recommend API calls with the Top-1 accuracy of 71.34%. Furthermore, MFLUTE can improve state-of-the-art approaches by up to 12 % in recommending accuracy.