Compatible Compositions Recommendation for Mashup-Oriented Depopularity Web API

With the increasing popularity of mashup development composed of various Web APIs in the realm of Web services, API recommendation research has gained significant attention. However, it faces challenges, particularly in dealing with nonpopular APIs. Studies suggest that focusing on nonpopular APIs may better align with developer needs, necessitating a deeper exploration of effective depopular API recommendation methods. Much of the existing research on nonpopular APIs recommendation employs deep learning methods, which focus on studying recommendation strategies to improve accuracy rather than composition recommendation strategies. Currently, there are two main challenges regarding depopular API composition recommendation methods, including compatibility concerns and the impact of popularity bias. To address these issues, this article proposes a novel approach called the depopularity web API compatible compositions recommendation approach (DPAR), which is a recommendation method designed for the compatible composition of Web APIs with long-tail mining capabilities that can mitigate the effects of popularity bias. DPAR builds a depopularity web API correlation graph (DP-ACG) and proposes a minimum weighted subgraph algorithm to select candidate Web API compositions that meet development requirements based on DP-ACG. The design of DP-ACG assures that candidate compositions generated by this method are depopularized and compatible. Subsequently, utilize the DPP algorithm to rerank the candidate Web API compositions, enhancing the diversity of the results. To evaluate DPARs performance, we conducted extensive experiments on real datasets to verify the efficacy of each module in DPAR and illustrate the impact of parameter settings. The experimental results demonstrate that DPAR consistently outperforms comparative methods, yielding significantly superior performance.