API Usage Recommendation Via Multi-View Heterogeneous Graph Representation Learning

Developers often need to decide which APIs use for the functions being implemented. With ever-growing number of and libraries, it becomes increasingly difficult developers find appropriate APIs, indicating necessity automatic API usage recommendation. Previous studies adopt statistical models or collaborative filtering methods mine implicit patterns However, they rely on occurrence frequencies mining patterns, thus prone fail low-frequency APIs. Besides, prior generally regard call interaction graph as homogeneous graph, ignoring rich information (e.g., edge types) in structure graph. In this work, we propose a novel method named MEGA improving recommendation accuracy especially Specifically, besides xmlns:xlink="http://www.w3.org/1999/xlink">call graph , MEGA considers another two new heterogeneous graphs: xmlns:xlink="http://www.w3.org/1999/xlink">global co-occurrence enriched with frequency xmlns:xlink="http://www.w3.org/1999/xlink">hierarchical project component information. three multi-view graphs, can capture more accurately. Experiments Java benchmark datasets demonstrate that significantly outperforms baseline by at least 19% respect Success Rate@1 metric. Especially, also increases baselines 55% regarding score.