Categorizing and Recommending API Usage Patterns Based on Degree Centralities and Pattern Distances

Abstract

Although efforts have been made on discovering and searching API usage patterns, how to categorize and recommend follow-up API usage patterns is still largely unexplored. This paper advances the state-of-the-art by proposing two methods for categorizing and recommending API usage patterns: first, categories of the usage patterns are automatically identified based on a proposed degree centrality-based clustering algorithm, and second, follow-up usage patterns of an adopted pattern are recommended based on a proposed metric of measuring distances between patterns. In the experimental evaluations, the patterns categorization can achieve 85.4% precision rate with 83% recall rate. The patterns recommendation had approximately half a chance of correctly predicting the follow-up patterns that were actually used by the programmers.