A Hybrid Pattern Knowledge Graph-Based API Recommendation Approach

There are a large number of application program interfaces (APIs) on the Internet. Due to frequently updating APIs, programmers are prompted to frequently consult API documents in the process of software development. However, the traditional query approaches of the documents have certain limitations. For example, the programmers need to know the API name as a prerequisite and are often unable to obtain the expected search results because of the difference in understanding between the description of the problem and the description of the documents. Only these "known-unknown" information can be queried, and it is difficult to query the "unknown-unknown" information. To address the limitations, we establish the knowledge graph of software source code combined with knowledge which is derived from the documents, Github project code warehouse, Stack overflow platform, and local project code warehouse. Moreover, we propose a hybrid pattern knowledge graph-based API recommendation approach for programmers to complete the query task of unknown-unknown information. Finally, we constructed large-scale real experiments. Evaluation results prove that the proposed approach significantly outperforms the state-of-the-art approach.